Depressive Symptoms Score Predicts Incident Type 2 Diabetes in Community Dwelling Old Icelandic People

Background

Depression is related to incident type 2 diabetes (T2D). However, little is known on this topic in older people from the Nordic countries and how health and lifestyle characteristics of participants affect this relationship. Thus, the aim of the present study was to investigate whether baseline depressive symptoms predict incident T2D in Icelandic older people and whether health and lifestyle characteristics of participants, can explain the relation between depression and diabetes.

Methods

We used data from the Age-Gene/Environment-Susceptibility-Reykjavik-Study (65-96 years). From the original sample of 3316 participants who finished follow-up, 2823 non-diabetic participants with a complete dataset on depressive symptoms and incident T2D at endpoint were included in this analysis. Depressive symptoms were assessed using the 15-item Geriatric Depression Scale (GDS).

Results

During a mean follow-up of 5.2 years, 103 (3.6%) of the 2823 participants developed T2D. According to the fully adjusted logistic regression model, baseline depressive symptoms in the highest category predicted incident T2D when compared to the lowest category (OR: 3.2; 95%CI: 1.3-8.2; p = 0.014). Statistical adjustment did only marginally alter the results. Subgroup analysis revealed that GDS was a significant predictor of incident T2D in most subgroups.

Conclusions

In older Icelandic people, having high depressive symptoms is a predictor of incident T2D during a follow-up period of 5.2 years. These associations are independent from health and lifestyle related covariates and are observed in most subgroups of our study population.

INTRODUCTION

Depression is a frequent and severe disease which can adversely impact feeling and thinking of an affected person and as a consequence disturb normal daily activities, e.g., sleeping, eating and working ¹. Depression is commonly observed in older people, however, prevalence numbers differ considerably between studies depending, among other things on cultural differences between populations and screening tools used in research. A recently published meta-analysis estimated pooled depression prevalence in individuals > 65 years to be 28% with the lowest numbers observed in Europe and the highest in Africa ². Further, women seem to be more frequently affected than men, but the sex difference in prevalence decreases with increasing age ³.

Depression in late life is a serious public health hazard, because it relates not only to low physical, cognitive and social performance, but also to higher risk of morbidity and suicide, which taken together results in increased mortality ^4-6. The rate of depression in people with diabetes mellitus is around two-fold higher when compared to healthy counterparts ⁷. Interestingly, the relationship between diabetes and depression seems to go both ways, as two meta-analyses indicate, depression is associated with increased odds of developing type 2 diabetes (T2D) by 18 to 60 percent ^8,9. Potential shared pathophysiological processes include inflammation which has been suggested to explain this relationship ¹⁰, as well as increased stress level by a hyperactivity in the hypothalamic-pituitary adrenal axis (HPA-axis) and sympathetic nervous system ^11,12. Further, a genetic overlap between T2D and depression has been suggested ¹³. It is also known that depression is associated with poor lifestyle, e.g., smoking ¹⁴, physical inactivity ¹⁵, and poor nutrition ¹⁶, factors known to play an important part in the aetiology of type 2 diabetes ^17,18.

The current epidemiological evidence on the associations between depression and incident type 2 diabetes is convincing ^8,9. However, to our best knowledge, no information on older people from the Nordic countries are available ^19,20. Also, many previous studies (reviewed in Mezuk et al., 2008; Graham et al., 2020, for single studies see Appendix 1) have not considered or only to limited extent the role of covariates as modulators of the relationship between depression and incident T2D.

Thus, the present paper studied the association between depressive symptoms and type 2 diabetes during 5.2 years of follow-up in community dwelling old Icelandic people using data from the prospective AGES-Reykjavik cohort study. The specific aims of the present study were to investigate whether baseline depressive symptoms predict incident T2D in older people; and whether other baseline characteristics of the participants, e.g., lifestyle and health, can explain the relation between depression and diabetes.

METHODS

STUDY POPULATION AND STUDY DESIGN

This longitudinal analysis is based on data from the AGES-Reykjavik study (n = 5764) enrolled in 2002-2006 as a continuation of the population-based Reykjavik Study (RS) in Iceland, initiated in 1967. Detailed baseline information has been described in a previous AGES-study paper ²¹. Between 2007-2011, AGES I participants returned to a second examination (58%, n = 3316), a 5-year follow-up visit (AGES II). The current study included participants who were not diabetic at baseline and had the relevant follow-up examination including information on incident type 2 diabetes (n = 2823).

ANTHROPOMETRICS

Weight and height were measured and BMI was calculated as kg/m². Body mass index was used as continuous variable.

MILD COGNITIVE IMPAIRMENT AND DEMENTIA

The criterion for MCI diagnosis was having deficits in memory or one other domain of cognitive function or deficits in at least 2 cognitive domains without being severe enough to cross the threshold for dementia and without loss of instrumental activities of daily living ²².

Assessment of dementia was done following a three-step protocol and according to international guidelines from the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition ²³. First, the digit symbol substitution test (DSST) ²⁴ and the Mini-Mental State Examination (MMSE) ²⁵ were administered to the total sample. Participants who scored 23 or lower on the MMSE or had a raw score of 17 or lower on the DSST were administered a second diagnostic cognitive test battery. Participants who scored 8 or more on Trails B ²⁶ (ratio of time taken for “Trails B/Trails A”) or had lower than total score of 19 for the four immediate recall trials of the Rey Auditory Verbal Learning ²⁷ went on to a third step. This step included a neurological test and a proxy interview regarding medical history, social, cognitive, and daily functioning changes of the participant.

DEPRESSIVE SYMPTOMS

The short form of the 15-item Geriatric Depression Scale (GDS) was used to estimate depressive symptoms. This questionnaire has been applied in healthy, medically ill and mild to moderately cognitively impaired older people. It has been extensively used in community, acute care, and long-term care settings. Four category of GDS were defined, scores of 0-4 as normal, 5-8 as mild depression; 9-11 as moderate depression; and 12-15 as severe depression ²⁸. In the present study sample, only 8.7% had a GDS score > 4, thus, GDS was categorized for statistical analysis as follows: 1^st category: GDS score = 0; 2^nd category: GDS score = 1-2; 3^rd category: GDS score = 3-4; 4^th category: GDS score => 4.

DIABETES MELLITUS AND PRE-DIABETES

Participants were categorized into having type 2 diabetis (diagnosed either as fasting serum glucose of ≥ 7 mmol/L, self-reported diabetes and/or use of diabetes medication), pre-diabetic (fasting blood glucose ≥ 5.5 to < 7 mmol/L) ²⁹ or normal health (neither of above definitions).

COVARIATES

Demographic and lifestyle data

Participants were asked about their age, gender, smoking (current smoking yes vs. no), alcohol consumption (yes vs no), marital status (single, divorced, widowed vs. married, cohabitation), fish oil consumption (< 2 times/week vs at least 3 times/week). Physical activity was assessed by a self-reported questionnaire and categorized as < 1 h/week, 1-3 h/week, and > 3 h/week. Education was categorized into two levels (elementary school or high school vs undergraduate or more than undergraduate education). Participants were instructed in advance to bring all medication they had used during the preceding two weeks before the clinic visit.

Laboratory data

The accredited IHA laboratory performed 25OHD measurements using unfrozen serum samples and the Liaison chemiluminescence immunoassay (DiaSorin Inc, Stillwater, Minnesota). Existing serum 25OHD levels were then standardized ³⁰. Glucose levels in a capillary blood sample were estimated by the Hoffman ferricyanide method, adapted to the Technicon-Method N-9a ³¹. Glucose was measured on a Hitachi 912, using reagents from Roche Diagnostics following the manufacturer’s instructions. Insulin was measured with a Roche Elecsys 2010 instrument ³².

STATISTICAL ANALYSIS

Statistical analyses were carried out using IBM SPSS version 26.0 (SPSS, Chicago, IL, USA). We used chi-square test for categorical variables and after visual inspection of the distribution, we used ANOVA or Kruskal Wallis test for continuous variables to test for statistical differences between GDS categories (Tab. I).

In order to calculate whether GDS categories status predict incident T2D (Tab. II) logistic regression analyses were applied controlling for various confounders. Model 1, the most basic model adjusted for age and gender; Models 2-6 adjusted in addition to age and gender for education and marital status (model 2), alcohol and smoking (model 3), number of medications and cognitive status (model 4), BMI and physical activity (model 5), fasting glucose and 25OHD (model 6). Model 7 was the fully adjusted model containing all of the above mentioned covariates. Identical models using GDS score instead of categories are shown in Appendix 2.

In order to investigate whether GDS predicts incident T2D similarly/differently in categories of subgroups of the study population, we used logistic regression in which T2D was the outcome variable and GDS score the main independent variable. The analyses were adjusted for age and gender. The following subgroups were evaluated: BMI (low/normal BMI, obesity), education (lower education, higher education), marital status (single/divorced/widowed, married/cohabitation), medications (0-4 medicines, at least 5 medicines), glucose metabolism (normal, prediabetes), sex (men, women), smoking (no, yes), physical activity(< 1 h/week, yes: 1-3 h/week, and > 3 h/week), alcohol (no, yes), 25OHD (below 50 nmol/L, at least 50 nmol/L), fish oil (< 2 times/week, at least 3 times/week) (Tab. III).

The level of statistical significance was set at p < 0.05.

RESULTS

The baseline characteristics of participants categorized by GDS score are shown in Table I. Of the participants (mean age = 75.0 ± 4.9 years), 555 (19.7%) had a GDS score = 0, 1401 (49.6%) had a GDS score = 1-2, 620 (22.0%) had a GDS score = 3-4, and 247 (8.7%) had a GDS score > 4. When comparing the four categories, there was the general tendency that the higher categories had more adverse or more disadvantageous characteristics in most of the variables measured at baseline. In accordance to that the need for at least 5 medications was 2.5 times more frequent in the highest GDS group compared to the lowest. Body mass index was however not significantly different between groups, and alcohol consumption was more frequent in the lower categories.

During a mean follow-up of 5.2 years, 103 (3.6%) of the 2823 participants developed T2D. The crude incidence T2D numbers for the GDS categories 1-4 were as follows:10 cases (1.8%), 42 cases (3.0%), 37 cases (6.0%) and 14 cases (5.7%), respectively.

Table II shows the results from logistic regression models estimating the incident T2D risk for the four GDS categories where the lowest category served as the reference. The minimally adjusted model 1 shows an around 3.5 times increased T2D risk for GDS categories 3 (p = 0.001) and 4 (p = 0.004) when compared to category 1, whereas the 1.7 times increased risk in category 2 was not significant. Further statistical correction for education and marital status (model 2), or alcohol and smoking (model 3), or number of medications and cognitive status (model 4), or BMI and physical activity (model 5), or fasting glucose and 25OHD (model 6) did attenuate the risk to a small degree, however, the risk difference between categories 3 and 4 vs 1 remained significant. In the fully corrected model 7, category 3 and 4 also remained significantly different from category 1. Besides GDS categories, BMI (OR = 1.06, p = 0.017), fasting glucose (OR = 1.16, p < 0.001) and number of medicines categories (OR = 1.57, p = 0.052) were predictors of T2D in the fully adjusted model 7 (numbers not shown in Table). Logistic models using GDS score instead of GDS categories draw a similar picture and indicate a significantly increased T2D risk by around 15% for a GDS increase by one. This increase was robust and independent from statistical correction (Appendix 2).

In the subgroup analysis we explored whether GDS score predicted incident T2D in various categories of a given subgroup in a similar way. The predictions for the categories within BMI, education, marital status, medication, smoking, alcohol, 25OHD and fish oil consumption were all similar. However, there were some numerical differences within the categories of glucose metabolism, sex and physical activity. Further analysis showed that interaction between GDS x sex and GDS x PA were not significant (P_interaction = 0.078 and 0.103, respectively), however, the interaction GDS x prediabetes was significant (P_interaction = 0.016).

DISCUSSION

The present study investigated the lon gitudinal associations between depressive symptoms and incident T2D in Icelandic old people from the AGES-Reykjavik cohort. The main result is that depressive symptoms are a strong predictor of T2D and mostly independent from other covariates. Additionally, we found the association to be robust and observed in most subgroups of the study population.

According to a recently published meta-analysis ⁹ using longitudinal studies of different designs, there is a positive relationship between depression and incident T2D. Different methods of identifying individuals with depression or depressive symptoms can be used to identify those who are at increased risk for T2D. Around two thirds of the studies included in this meta-analysis were from the USA, the rest mainly from Europe and Asia. Two studies from the Nordic countries ^19,20 included participants from young adulthood, however, no information on older people from the Nordic countries was included. Participants in the present study had a mean age of 75 years and displayed low mean levels of depressive symptoms. The results are largely in agreement with the results from this above mentioned meta-analysis. We observed a higher OR, i.e., 3.2 (highest vs lowest GDS category in the fully adjusted model) in comparison to the mean estimate of 1.18 from the meta-analysis. Years of follow-up, identification of depression and T2D were all comparable. Previous meta-analyses have also indicated somewhat higher risk estimates, i.e., 1.32 to 1.60 ^8,33.

Several possible explanations for the link between depression and T2D have been proposed ^10-13. Individuals with depression have often poorer health related characteristics and lifestyle which can be connected to future risk of T2D and could thus explain or at least modulate the association between depression and incident T2D ¹⁰. Our study results support these finding from previous studies in a way that our participants in the higher GDS categories were more often smokers, physically inactive and had lower circulating vitamin D when compared to lower categories. Further, they tended to be older, took more medications, had lower MMSE score and had more often MCI or dementia.

However, our statistical analysis does not indicate that health and lifestyle differences at baseline between the GDS categories explain the increased T2D risk of the higher GDS category in the longitudinal setting. Contrary to our expectations, extensive covariate adjustment in the logistic regression models did change the estimated OR marginally at most. Residual confounding, which is considered to be confounding that persists despite statistical correction due to imperfect measurements and/or lack of available covariate data ³⁴, might explain partly our findings. However, our data also support the hypothesis that there might be a causal link between depression itself and T2D.

A possible common causal link could be increased level of chronic stress ¹⁰ enforcing the HPA-axis and the sympathetic nervous system (SNS), by the elevated secretion of cortisol, adrenalin and noradrenalin ³⁵. Prolonged elevation of these hormones potentially disturbs glucose metabolism, promotes body fat aggregation and could thus result in T2D. Further, chronic stress impacts risk of depression by activation of the fear system and a decrease in response of the reward system ³⁶. It has also been suggested that chronic stress affects the balance of the immune system by elevating the secretion of pro-inflammatory cytokines, which may interact with pancreatic β-cells, induce insulin resistance and result in T2D. Newer research has reported that inflammation plays also part in the development of depression by altered neurotransmitter metabolism, neuroendocrine function, synaptic plasticity and behaviour ^37-39.

Further, although some previous studies did not indicate that genetic characteristics explain the association between depression and T2D ^40,41, a more recently published study using data from Swedish and Danish twin registries reported that in both populations, the association between depression and T2D diabetes were explained by genetic effects to a degree ¹³.

A subgroup analysis of our result shows that the association between baseline depressive symptoms and incident T2D is a robust finding and can be observed in most of the investigated subgroups. Although the association was not significant in all subgroups, the general direction and strength of the association was mostly similar, and differences in p-value can be attributed to differences in statistical power due to differences in sample size of the subgroups. Nevertheless, there is considerable numerical difference in the OR between the subgroups of glucose metabolism, i.e., normal and prediabetes. While participants with normal glucose metabolism had 36% increased risk (OR: 1.36, p < 0.001) for having T2D the risk was only 10% increased (OR: 1.10, p = 0.046) among prediabetes. Interestingly, a study by Rubin et al. ⁴² who investigated subjects at high-risk for diabetes did not find depressive symptoms to be a predictor of incident T2D. Possibly, the deteriorated metabolism in prediabetic participants results in such a great diabetes risk, that an increased GDS score does no longer play as strong role in the risk for incident T2D in this subgroup.

STRENGHTS AND LIMITATIONS

The analyses presented in this paper are based on data from the AGES Reykjavik study, which was a prospective cohort study in Icelandic old people. It is a strength of this cohort, that it measured a great number of relevant covariates thus allowing the investigation of the association between depressive symptoms and incident T2D with consideration of possible interactions of health and lifestyle characteristics of the participants.

However, it is a limitation of the present study that the distribution of depressive symptoms measured by GDS was limited. When following the categorization suggested by the authors of GDS, only 8.7% of the participants were considered to have at least mild depressive symptoms. However, re-categorization of the participants on bases of actual statistical distribution yielded categories of meaningful sample size, thus allowing statistical analysis.

It is a further limitation of the current study that information on use of antidepressant medications was not available for the data analysis. Although it has been shown that different methods of identifying subjects with depression yield similar results in term of T2D prediction ⁹, a double approach where we would have investigated both the association between GDS and T2D as well as use of antidepressant medications and T2D would have further informed our results.

CONCLUSIONS

Our study shows that in older Icelandic people, depressive symptoms are a strong predictor of incident T2D during a follow-up period of 5.2 years. This association is independent from most of the covariates and observed in most subgroups of the study population.

Acknowledgments

This work was supported by The Foundation of St. Josef’s Hospital in cooperation with The Icelandic Gerontological Research Center, National University Hospital. The funding sources did not have any role in the study design, conduct of the study, analysis of the data, or manuscript preparation.

Conflict of interest statement

The authors declare no conflict of interest.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Author contributions

AR, HE: contributed to the design and conceptualization of the study, contributed to the analysis, interpretation of the data and drafted the manuscript; MC, VG: provided critical overview of study analysis and interpretation. All authors provided critical revisions and acceptance of the final version of this scientific manuscript.

Ethical consideration

The study was approved by the National Bioethics Committee in Iceland (approval number VSN-00-063), the Data Protection Authority and by the National Institute on Aging Intramural Institutional Review Board. Written informed consent was obtained from all participants.

Figures and tables

Table I. Baseline characteristics of the participants according to Geriatric Depression Scale categories (n = 2823).

	1st category			2nd category			3rd category			4th category
	(n = 555)			(n = 1401)			(n = 620)			(n = 247)			p-value
Age (years)	74.0	±	4.6	75.0	±	4.8	75.8	±	5.1	75.8	±	5.1	< 0.001
Men (%)		18.3			51.7			22.6			7.4		0.058
Women (%)		20.6			48.2			21.5			9.6
Higher education (%)		35.6			27.6			24.0			21.8		< 0.001
Married/cohabitation (%)		65.3			65.4			61.2			54.3		0.004
Smoking (yes in %)		6.7			7.3			12.0			11.0		< 0.001
Vigorous exercise > 3 h/week (%)		28.1			18.6			11.2			8.6		< 0.001
Alcohol (yes in %)		74.3			68.3			70.0			58.8		< 0.001
BMI (kg/m2)	26.8	±	3.9	27.1	±	4.2	27.3	±	4.1	27.2	±	4.3	0.309
Glucose (mg/dL)	5.5	±	0.5	5.5	±	0.5	5.5	±	0.5	5.4	±	0.5	0.017
Insulin (uIU/mL)	1.2	±	0.8	1.4	±	0.9	1.5	±	1.0	1.4	±	0.9	< 0.001
25 OH-vitamin D (nmol/L)	60.2	±	16.7	59.7	±	17.5	56.8	±	17.6	54.9	±	19.0	< 0.001
Medication (at least 5)		20.4			29.2			36.0			49.8		< 0.001
MMSE (score)	27.8	±	1.8	27.3	±	2.4	26.8	±	2.6	26.7	±	2.5	< 0.001
Mild cognitive impairment (%)		2.9			5.3			9.2			10.7		< 0.001
Dementia (%)		0.7			0.7			2.1			3.7		< 0.001
Data are presented as the mean ± SD for continuous variables and as % for categorical variables. P-value based on one-way-ANOVA/ Kruskal Wallis test (continous variables) and chi-squared test (categorical variables).
Geriatric Depression Scale (GDS) categories: 1st category: GDS score = 0; 2nd category: GDS score = 1-2; 3rd category: GDS score = 3-4; 4 category: GDS score => 4.

Table I. Baseline characteristics of the participants according to Geriatric Depression Scale categories (n = 2823).

	1st category			2nd category			3rd category			4th category
	(n = 555)			(n = 1401)			(n = 620)			(n = 247)			p-value
Age (years)	74.0	±	4.6	75.0	±	4.8	75.8	±	5.1	75.8	±	5.1	< 0.001
Men (%)		18.3			51.7			22.6			7.4		0.058
Women (%)		20.6			48.2			21.5			9.6
Higher education (%)		35.6			27.6			24.0			21.8		< 0.001
Married/cohabitation (%)		65.3			65.4			61.2			54.3		0.004
Smoking (yes in %)		6.7			7.3			12.0			11.0		< 0.001
Vigorous exercise > 3 h/week (%)		28.1			18.6			11.2			8.6		< 0.001
Alcohol (yes in %)		74.3			68.3			70.0			58.8		< 0.001
BMI (kg/m2)	26.8	±	3.9	27.1	±	4.2	27.3	±	4.1	27.2	±	4.3	0.309
Glucose (mg/dL)	5.5	±	0.5	5.5	±	0.5	5.5	±	0.5	5.4	±	0.5	0.017
Insulin (uIU/mL)	1.2	±	0.8	1.4	±	0.9	1.5	±	1.0	1.4	±	0.9	< 0.001
25 OH-vitamin D (nmol/L)	60.2	±	16.7	59.7	±	17.5	56.8	±	17.6	54.9	±	19.0	< 0.001
Medication (at least 5)		20.4			29.2			36.0			49.8		< 0.001
MMSE (score)	27.8	±	1.8	27.3	±	2.4	26.8	±	2.6	26.7	±	2.5	< 0.001
Mild cognitive impairment (%)		2.9			5.3			9.2			10.7		< 0.001
Dementia (%)		0.7			0.7			2.1			3.7		< 0.001
Data are presented as the mean ± SD for continuous variables and as % for categorical variables. P-value based on one-way-ANOVA/ Kruskal Wallis test (continous variables) and chi-squared test (categorical variables).
Geriatric Depression Scale (GDS) categories: 1st category: GDS score = 0; 2nd category: GDS score = 1-2; 3rd category: GDS score = 3-4; 4 category: GDS score => 4.

Table II. Associations between GDS and risk of type 2 diabetes among AGES-Reykjavik participants (n = 2823).

Models	GDS categories	OR	95%CI		p-value
Model 1	1st category	1.000			ref.
	2nd category	1.679	0.835	3.376	0.146
	3rd category	3.489	1.711	7.113	0.001
	4th category	3.378	1.474	7.744	0.004
Model 2	1st category	1.000			ref.
	2nd category	1.663	0.826	3.346	0.154
	3rd category	3.428	1.678	7.002	0.001
	4th category	3.299	1.436	7.582	0.005
Model 3	1st category	1.000			ref.
	2nd category	1.667	0.829	3.354	0.152
	3rd category	3.541	1.734	7.231	0.001
	4th category	3.351	1.457	7.707	0.004
Model 4	1st category	1.000			ref.
	2nd category	1.585	0.786	3.193	0.198
	3rd category	3.198	1.561	6.553	0.001
	4th category	2.917	1.257	6.768	0.013
Model 5	1st category	1.000			ref.
	2nd category	1.510	0.745	3.062	0.253
	3rd category	3.061	1.481	6.328	0.003
	4th category	2.998	1.286	6.988	0.011
Model 6	1st category	1.000			ref.
	2nd category	1.411	0.679	2.931	0.357
	3rd category	2.881	1.355	6.126	0.006
	4th category	3.953	1.622	9.632	0.002
Model 7	1st category	1.000			ref.
	2nd category	1.275	0.608	2.675	0.520
	3rd category	2.605	1.201	5.646	0.015
	4th category	3.236	1.270	8.244	0.014
OR: Odds Ratio. GDS: Geriatric Depression Scale. Based on logistic regression models.
For all models:
1st category: GDS score = 0; n = 555, 10 cases = 1.8%
2nd category: GDS score = 1-2; 1401, 42 cases = 3.0%
3rd category: GDS score = 3-4; n = 620, 37 cases = 6.0%
4th category: GDS score = >4; n = 247, 14 cases = 5.7%
Model 1: corrected for age and sex
Model 2: corrected for age, sex, education and marital status
Model 3: corrected for age, sex, alcohol and smoking
Model 4: corrected for age, sex, number of medications and cognitive status
Model 5: corrected for age, sex, BMI and physical activity
Model 6: corrected for age, sex, fasting glucose and 25OHD
Model 7: corrected for age, sex, education, marital status, alcohol, smoking, number of medications, cognitive status, BMI, physical activity, fasting glucose and 25OHD

Table II. Associations between GDS and risk of type 2 diabetes among AGES-Reykjavik participants (n = 2823).

Models	GDS categories	OR	95%CI		p-value
Model 1	1st category	1.000			ref.
	2nd category	1.679	0.835	3.376	0.146
	3rd category	3.489	1.711	7.113	0.001
	4th category	3.378	1.474	7.744	0.004
Model 2	1st category	1.000			ref.
	2nd category	1.663	0.826	3.346	0.154
	3rd category	3.428	1.678	7.002	0.001
	4th category	3.299	1.436	7.582	0.005
Model 3	1st category	1.000			ref.
	2nd category	1.667	0.829	3.354	0.152
	3rd category	3.541	1.734	7.231	0.001
	4th category	3.351	1.457	7.707	0.004
Model 4	1st category	1.000			ref.
	2nd category	1.585	0.786	3.193	0.198
	3rd category	3.198	1.561	6.553	0.001
	4th category	2.917	1.257	6.768	0.013
Model 5	1st category	1.000			ref.
	2nd category	1.510	0.745	3.062	0.253
	3rd category	3.061	1.481	6.328	0.003
	4th category	2.998	1.286	6.988	0.011
Model 6	1st category	1.000			ref.
	2nd category	1.411	0.679	2.931	0.357
	3rd category	2.881	1.355	6.126	0.006
	4th category	3.953	1.622	9.632	0.002
Model 7	1st category	1.000			ref.
	2nd category	1.275	0.608	2.675	0.520
	3rd category	2.605	1.201	5.646	0.015
	4th category	3.236	1.270	8.244	0.014
OR: Odds Ratio. GDS: Geriatric Depression Scale. Based on logistic regression models.
For all models:
1st category: GDS score = 0; n = 555, 10 cases = 1.8%
2nd category: GDS score = 1-2; 1401, 42 cases = 3.0%
3rd category: GDS score = 3-4; n = 620, 37 cases = 6.0%
4th category: GDS score = >4; n = 247, 14 cases = 5.7%
Model 1: corrected for age and sex
Model 2: corrected for age, sex, education and marital status
Model 3: corrected for age, sex, alcohol and smoking
Model 4: corrected for age, sex, number of medications and cognitive status
Model 5: corrected for age, sex, BMI and physical activity
Model 6: corrected for age, sex, fasting glucose and 25OHD
Model 7: corrected for age, sex, education, marital status, alcohol, smoking, number of medications, cognitive status, BMI, physical activity, fasting glucose and 25OHD

Table III. Associations between GDS score (predictor variable) and incident T2D ⁴³ categorized by different subgroups.

Subgroups	Categories within subgroups	N †	T2D cases ‡	OR §	95%CI		p-value
BMI	Low/normal BMI	2211	61	1.14	1.03	1.26	0.014
	Obesity	612	42	1.15	1.01	1.32	0.039
Education	Lower education	2038	78	1.15	1.05	1.25	0.002
	Higher education	785	25	1.13	0.92	1.38	0.248
Marital status	Single, divorced, widowed	1029	38	1.13	1.00	1.29	0.050
	Married, cohabitation	1794	65	1.15	1.04	1.28	0.007
Medication	0-4 medicines/day	1955	56	1.12	0.98	1.27	0.099
	At least 5 medicines	868	47	1.13	1.01	1.25	0.027
Glucose metabolism	Normal	1641	15	1.36	1.17	1.58	< 0.001
	Prediabetes	1182	88	1.10	1.00	1.21	0.046
Sex	Men	1142	49	1.04	0.90	1.20	0.633
	Women	1681	54	1.21	1.10	1.34	< 0.001
Smoking	No	2583	96	1.14	1.05	1.24	0.003
	yes	340	7	1.23	0.97	1.56	0.082
Physical activity	No	1102	50	1.10	0.99	1.23	0.088
	Yes	1721	53	1.17	1.04	1.32	0.008
Alcohol	No	872	36	1.09	0.96	1.23	0.192
	Yes	1951	67	1.19	1.07	1.32	0.001
25OHD	Below 50 nmol/L	845	35	1.07	0.92	1.23	0.382
	At least 50 nmol/L	1978	68	1.19	1.08	1.31	< 0.001
Fish oil	< 2 times/week	1172	48	1.11	0.97	1.25	0.120
	> 3 times/week	1651	55	1.17	1.06	1.30	0.002
OR: Odds Ratio; GDS: Geriatric Depression Scale; 25OHD: 25 hydroxyvitamin D; BMI: body mass index; T2D: type 2 diabetes. Based on logistic regression adjusted for age and sex (analysis on sex category was only adjusted for age). Each line in the table represents a logistic regression for a given subgroup.
† Number of participants in each category (in each line).
‡ Number of T2D cases in each category (in each line).
§ OR shows the the estimated change in risk in incident T2D if GDS score (continious) increases by 1.

Table III. Associations between GDS score (predictor variable) and incident T2D ⁴³ categorized by different subgroups.

Subgroups	Categories within subgroups	N †	T2D cases ‡	OR §	95%CI		p-value
BMI	Low/normal BMI	2211	61	1.14	1.03	1.26	0.014
	Obesity	612	42	1.15	1.01	1.32	0.039
Education	Lower education	2038	78	1.15	1.05	1.25	0.002
	Higher education	785	25	1.13	0.92	1.38	0.248
Marital status	Single, divorced, widowed	1029	38	1.13	1.00	1.29	0.050
	Married, cohabitation	1794	65	1.15	1.04	1.28	0.007
Medication	0-4 medicines/day	1955	56	1.12	0.98	1.27	0.099
	At least 5 medicines	868	47	1.13	1.01	1.25	0.027
Glucose metabolism	Normal	1641	15	1.36	1.17	1.58	< 0.001
	Prediabetes	1182	88	1.10	1.00	1.21	0.046
Sex	Men	1142	49	1.04	0.90	1.20	0.633
	Women	1681	54	1.21	1.10	1.34	< 0.001
Smoking	No	2583	96	1.14	1.05	1.24	0.003
	yes	340	7	1.23	0.97	1.56	0.082
Physical activity	No	1102	50	1.10	0.99	1.23	0.088
	Yes	1721	53	1.17	1.04	1.32	0.008
Alcohol	No	872	36	1.09	0.96	1.23	0.192
	Yes	1951	67	1.19	1.07	1.32	0.001
25OHD	Below 50 nmol/L	845	35	1.07	0.92	1.23	0.382
	At least 50 nmol/L	1978	68	1.19	1.08	1.31	< 0.001
Fish oil	< 2 times/week	1172	48	1.11	0.97	1.25	0.120
	> 3 times/week	1651	55	1.17	1.06	1.30	0.002
OR: Odds Ratio; GDS: Geriatric Depression Scale; 25OHD: 25 hydroxyvitamin D; BMI: body mass index; T2D: type 2 diabetes. Based on logistic regression adjusted for age and sex (analysis on sex category was only adjusted for age). Each line in the table represents a logistic regression for a given subgroup.
† Number of participants in each category (in each line).
‡ Number of T2D cases in each category (in each line).
§ OR shows the the estimated change in risk in incident T2D if GDS score (continious) increases by 1.