Introduction

As early as the 17th century, Thomas Willis, the first physician to write the English medical treatise about diabetes mellitus, mentioned that ‘nervous juice hurtful to other humors and prolonged sorrow appeared to be important risk factors for diabetes mellitus’ [1]. Over the past 20 years, a body of studies has investigated associations between adverse psychosocial factors and diabetes mellitus development and prognosis. Exposure to psychosocial stress is associated with a number of pathophysiological mechanisms that make an association between stress and diabetes risk theoretically plausible. For example, psychosocial stress can stimulate the hypothalamus–pituitary–adrenal axis, the sympathetic nervous system and inflammatory pathways known to affect glucose metabolism [2, 3]. However, evidence from clinical trials on the efficacy of psychological interventions for the treatment of diabetes remains equivocal [4, 5]. The several reviews on this topic have reported disparate findings. This may be because some reviews have not distinguished between prospective studies and cross-sectional or retrospective case–control studies [68]. Cross-sectional and retrospective case–control studies are subject to recall bias caused by diabetes mellitus diagnosis or memory distortion, and cannot conclusively detect a longitudinal association between predictors and outcome variables. Furthermore, none of the previous reviews have used meta-analytic techniques to quantify the extent to which adverse psychosocial factors affect diabetes mellitus [610].

Psychosocial stress can be considered a product of exposure to a stressor and the human response to it. Thus, several factors are relevant to stress responses, including cognitive appraisals, behavioural coping and the use of social support [3, 11]. The conceptualisation of social support covers instrumental support, including both financial and assistance with tasks, emotional/appraisal support, information, companionship and self-esteem support [12]. Accordingly, several psychosocial categories have been defined: (1) stressful events (e.g. life events, job stress, severe chronic stress and daily stress); (2) stress-prone personality or coping style (e.g. avoidant coping, denial coping, neuroticism, hopelessness); and (3) poor social support (e.g. poor social participation, poor stable partnership, poor family contact, loneliness).

The aim of this article is to conduct a systematic review and meta-analysis of prospective cohort studies to explore and quantify the putative causal associations of adverse psychosocial factors with the development and prognosis of diabetes mellitus. The following questions will be addressed: (1) What adverse psychosocial factors are associated with diabetes mellitus? (2) Do associations with adverse psychosocial factors differ according to methodological study quality, follow-up periods and participant characteristics (age and type 1 or type 2 diabetes)? The roles of behavioural and biological pathways in the association between adverse psychosocial factors and diabetes mellitus are also discussed. Given that the disease burden of diabetes can in itself be considered a powerful chronic stressor, we hypothesised that the association between adverse psychosocial factors and diabetes prognosis would be stronger than the association with diabetes development.

Methods

Data sources and searches

We developed a protocol using a widely recommended method for conducting systematic reviews of observational studies [13]. We searched the general bibliographic databases Medline (1966–July 2008), PsycINFO (1872–July 2008), Web of Science (1900–July 2008) and PubMed (1950–July 2008), and scrutinised reference lists from relevant reviews and articles. The main search strategy was diabetes mellitus AND (psychological stress OR psychosocial stress OR work stress OR life event OR life stress OR chronic stress OR social support OR personality OR coping) AND (prospective OR longitudinal).

Study selection

We limited the current systematic review and meta-analysis to prospective studies. Cross-sectional and retrospective case–control studies are subject to recall bias and cannot conclusively identify the temporal association between predictors and outcome variables. As mentioned in the introduction, adverse psychosocial factors were divided into the following categories: (1) stressful events; (2) stress-prone personality or coping style; and (3) poor social support. Criteria for inclusion were as follows: (1) English language full-length publication in a peer-reviewed journal; (2) prospective cohort design; (3) investigation of a longitudinal association of adverse psychosocial factors with diabetes mellitus control or incidence. If more than one kind of adverse psychosocial factor was assessed in one paper, the samples were included separately. If more than one type of diabetes mellitus control was assessed in one paper, the data for the longer term indicator of glycaemic control were included (e.g. HbA1c rather than average self-monitoring blood glucose levels for a week). Studies that used low socioeconomic status (SES) as an indicator for adverse psychosocial factors were excluded, because almost all studies included SES as a covariate, and low SES alone has many consequences other than greater stress experience [14, 15]. Studies that examined the association between diabetes and psychiatric illnesses or psychological distress, such as depression and anxiety, were excluded because there may be illness-related disturbances of physiological function in these patients [16]. If a cohort was analysed in more than one publication, the paper with shorter follow-up, smaller sample size or poorer study quality was excluded.

Data extraction and quality assessment

A manual was prepared for coding the studies. The manual was revised during the coding to incorporate important aspects of the identified studies. The final list of variables included first author and publication year, cohort size with participant characteristics (country), follow-up duration (years), type of adverse psychosocial factors (method of measurement), controlled covariates, diabetes outcome (method of measurement), quality score and brief results/effect size (correlation coefficient [r] in diabetes control, HR or RR with 95% CIs in diabetes incidence). The selected studies were categorised into two groups: associations between adverse psychosocial factors and diabetes control in diabetic patients; and associations between adverse psychosocial factors and diabetes incidence in a population without diabetes at enrolment. Categorising the studies in this manner enabled us to examine both the prognostic and aetiological effects of adverse psychosocial factors on diabetes. When primary sources provided insufficient data, we obtained additional information by other means, such as communication with the author or from indirect calculations.

We assessed all manuscripts for quality, since this can contribute to the potential bias associated with the effect estimation. In line with previous studies [1719], we considered a study to be of good quality if (1) participants were recruited consecutively or randomly or a representative population was used; (2) explanatory variables were ascertained by validated instruments or clinical examination; (3) outcome variables were ascertained by validated instruments or clinical examination; and (4) possible covariates were controlled for, including age, sex, smoking, alcohol consumption, BMI or physical activity level, and SES, and for populations with diabetes mellitus, basal disease status and medical treatment. These quality criteria were scored one point each. Studies were placed into high or low quality categories according to whether or not they fulfilled three or more of these criteria.

Study inclusion and data extractions were conducted by one author (Y. Chida) and verified by another (M. Hamer). Quality and validity were assessed independently by at least two reviewers. Disputes were settled by consensus.

Data synthesis and analysis

Meta-analytic procedures were followed that have been described elsewhere [20, 21]. Briefly, an effect size was calculated from the difference between the diabetes and control groups or the incidence between the control and exposed groups. In the case of association between adverse psychosocial factors and diabetes control, this was then transformed into r, which gives more weight to larger studies, since they tend to obtain more reliable estimates of the population effect size. If the raw data were not presented, then an F ratio for the main effect over time was used for conversion into r instead. If no relevant convertible statistics were presented, other than a p value, we calculated the t statistic from the p value and an r-sub (equivalent) [22]. When a paper reported p < 0.05, p < 0.10 or NS, we computed r-sub (equivalent) with p values of 0.025, 0.050, 0.50 (one-tailed), respectively, which likely yielded a highly conservative estimate of the effect size. We performed random effects modelling [23] in our analyses, which takes into account the amount of variance caused by differences between studies as well as differences among participants within studies. Therefore, an inferential statistic or effect size was used to calculate a z score for each study, which was weighted by sample size and then modified by the estimated between-study variance. In the case of association between adverse psychosocial factors and incident diabetes, HRs or RRs were calculated as measures of effect size. In each case, HRs or RRs were logarithmically (loge, ln) transformed, and standard errors and corresponding 95% CIs were calculated from ln(HR) or ln(RR). Meta-analyses normally include only one effect size per construct per study. In the present meta-analysis, the concept of construct was defined such that each psychosocial category measure was viewed as a separate construct. Thus, several effect sizes were often derived from a given study, each representing a different construct. When more than one measure was reported for a particular construct (e.g. several measures of stressful events), the unweighted mean of each set of effect sizes (e.g. mean r for all stressful events) was used to calculate the effect size. Separate meta-analyses were carried out for each of the three categories of psychosocial factor for the diabetes mellitus control and incidence. If more than one psychosocial factor was analysed in a single article, they were included as separate studies. Provided there was sufficient information (three or more studies), we aimed to perform sensitivity analyses according to the characteristics of the study population (age and type 1 or type 2 diabetes), methodological study quality and type of adverse psychosocial factors. We simultaneously employed the I 2 statistic for homogeneity between studies, which indicates the proportion of the total variation across studies that is not explained by chance [24]. Finally, to detect publication biases, we measured the degree of asymmetry by using Egger’s unweighted regression asymmetry test [25]. All analyses were performed using the Meta-Analysis Program [26].

Results

A flow diagram detailing the number of studies selected for the present systematic review is shown in Fig. 1. Tables 1 and 2, and the Electronic supplementary material (ESM Table 1), detail the papers that were included in (n = 14) [2740] and excluded from (n = 11) the review. There were 16 relationships between psychosocial measures and disease-related factors in populations with diabetes and six relationships between psychosocial measures and disease-related factors in populations without diabetes at baseline.

Fig. 1
figure 1

Flow diagram of systematic review (The Quality of Reporting of Meta-analyses [QUOROM] statement flow diagram)

Table 1 Prospective studies investigating the associations of adverse psychosocial factors with diabetes control in diabetic populations
Table 2 Prospective studies investigating the associations of adverse psychosocial factors with diabetes incidence in populations without any diagnosed diabetes

Study characteristics and quality

Results from 11 diabetes cohorts and three cohorts without any diagnosed diabetes have been published over the two decades between 1987 and 2008, involving participants from a wide range of countries in Asia, Australasia, Europe and America. The studies of populations without any diagnosed diabetes involved larger samples and a higher proportion of these studies had longer follow-up periods compared with the diabetes studies (sample size ± SD: 7,528.3 ± 9,477.7 vs 94.0 ± 81.2, respectively; ≥2 years follow-up: 100% vs 50%, respectively). Stressful events were principally evaluated in the studies of populations without any diagnosed diabetes. By contrast, other categories of adverse psychosocial factors were assessed more frequently in diabetes studies than in populations without any diagnosed diabetes. Regarding the type of diabetes, diabetes mellitus population studies estimated type 1 more frequently than type 2 diabetes, while the population without any diagnosed diabetes included only type 2 diabetes. The study quality score (0–4) of the diabetes and non-diabetes studies averaged 2.5 ± 0.5 and 2.3 ± 0.5, respectively.

Study results and meta-analysis

The overall combined r value was 0.096 (95% CI 0.028–0.163, p = 0.006) for the diabetes studies (Fig. 2) and the overall combined HR was 1.03 (95% CI 0.93–1.13, p = 0.61) for the populations without any diagnosed diabetes, suggesting a harmful effect of adverse psychosocial factors on diabetes control in the diabetic population. Notably, the overall finding in diabetic populations was not accompanied by publication bias (see Fig. 3). The total variation across studies was 25% (95% CI 26–61%).

Fig. 2
figure 2

Forest plots of individual studies investigating the association between adverse psychological factors and diabetes mellitus. Individual study symbols are proportional in size to the weight of the study. Only those studies for which effect sizes could be computed have been included

Fig. 3
figure 3

Funnel plots depicting the relationship between adverse psychological factors and diabetes mellitus. Overall association between adverse psychological factors and diabetes in diabetic populations

As shown in Fig. 4, the subgroup meta-analyses on diabetes studies with longer follow-up periods (≥2 years) and younger people (≤18 years old) showed significant combined r values (0.217, CI 0.093–0.3334; and 0.129, CI 0.007–0.248, respectively). However, the diabetes studies with higher quality scores (≥3) or larger sample sizes (≥70) demonstrated slightly reduced combined r values compared with the overall effect size. Unfortunately, there were insufficient studies to carry out further subanalyses for populations without any diagnosed diabetes.

Fig. 4
figure 4

Results of meta-analyses, subgrouping, and sensitivity analyses. Higher positive correlation coefficients indicate a more detrimental role of stress in diabetes control

Differences were observed between the adverse psychosocial factor categories. In the diabetes studies, poor social support was more strongly associated with a poorer outcome compared with the overall effect (0.149, CI 0.066–0.231). In contrast, the associations between stressful events and diabetes and between stress-prone personality or coping style and diabetes mellitus were no longer significant. The analyses of diabetes type demonstrated that, when analysed together, type 1 and type 2 diabetic populations exhibited similar associations with adverse psychosocial factors, while type 1 diabetes alone did not show significant associations with adverse psychosocial factors.

Discussion

The present investigation is the first quantitative systematic review to show that adverse psychosocial factors are significantly associated with poorer control of type 2 diabetes and tend to exacerbate type 1 diabetes in participants with existing diabetes. There was no association between adverse psychosocial factors and diabetes in cohorts without diabetes at baseline, although this finding should be interpreted with caution owing to limited studies. Intriguingly, poor social support was more strongly associated with poor diabetes mellitus control rather than stressful events per se and stress-prone personality or coping style.

Possible underlying mechanisms

The harmful relationship between adverse psychosocial factors and subsequent diabetes control might be primarily mediated via behavioural pathways, although we were unable to carry out a subgroup analysis because only two studies adjusted for major behavioural covariates (smoking, alcohol consumption, BMI or physical activity level and SES). For example, adverse psychosocial factors might promote high-risk behaviours, such as poor diet, less physical activity, smoking, excessive alcohol consumption, poor sleep and lower treatment adherence [10]. Another possible explanation is that direct physiological pathways are involved in associations between adverse psychosocial factors and poor diabetes control. Adverse psychosocial factors stimulate the hypothalamus–pituitary–adrenal axis and the sympathetic nervous system, resulting in increased release of cortisol, adrenaline (epinephrine) and neuropeptide Y. Cortisol is a stress hormone that triggers glucose production, increases lipolysis and circulating NEFA, decreases insulin secretion from beta cells and decreases sensitivity to insulin [41, 42]. Adrenaline has similar effects to cortisol on glucose and fat metabolism [41, 42], and neuropeptide Y may mediate stress-induced obesity and the metabolic syndrome by increasing adipogenesis and lipolysis [43], thereby resulting in poor control of diabetes.

Issues arising from the sensitivity analyses

The subgroup analyses showed slightly reduced effects in studies with larger sample sizes compared with the overall effect, although those with longer follow-up periods exhibited stronger effects. Generally, the cohort studies with larger sample sizes and longer follow-up periods are considered to provide stronger evidence, because these designs increase the power to detect any differences between the control and exposed groups. Furthermore, study quality is important [44], and the subgroup analysis of studies with a high quality score (≥3) showed a slightly reduced association between adverse psychosocial factors and poor diabetes control compared with the overall association. Nevertheless, in the present meta-analyses, we found no evidence of publication bias—a positive result bias is obtained if authors are more likely to submit, or editors accept, positive than null (negative or inconclusive) results.

The subgroup meta-analyses across the three different types of adverse psychosocial factors found that low social support was more strongly associated with poorer diabetes control than were stressful events and stress-prone personality or coping style. Given that the disease burden of diabetes can in itself be considered a powerful chronic stressor, this may explain why factors such as social support are important predictors of diabetes control. Randomised controlled trials have shown that psychological interventions have an efficient effect on glycaemic control in patients with type 1 or type 2 diabetes [4, 5], which suggests that organising social support may be rather more important than preventing stressful events in relation to diabetes or managing human cognition, appraisal and coping. It is also interesting that adverse psychosocial factors predicted diabetes control in younger populations (≤18 years old), suggesting that young individuals are more likely to respond to adverse psychosocial factors. In line with this finding, the meta-analysis conducted by Winkley et al. [5] showed that psychological interventions are effective in children and adolescents, but not adults. More intriguingly, the analyses of diabetes types found that adverse psychosocial factors were more robustly associated with poor control in type 2 than type 1 diabetes, although possible explanations of this difference remain to be identified.

Limitations and guidelines for future studies

The aetiological association of adverse psychosocial factors with diabetes remains undetermined at present because only two cohorts have been published. However, a recent meta-analysis of nine prospective studies reported that depressed adults have a 37% increased risk of developing type 2 diabetes [45]. Negative life events during the first 2 years of life, such as divorce, high parenting stress and foreign origin of the mother predicted an increased risk of diabetes-related autoimmunity, which critically contributes to the development of type 1 diabetes [46, 47]. In addition, inflammatory cytokines such as IL-6 are known to increase in response to acute psychosocial stress [48], and these inflammatory markers have also been prospectively associated with an increased risk of diabetes among a large cohort of postmenopausal women [49]. Taken together, these findings indicate that a relationship between adverse psychosocial factors and the development of diabetes is feasible.

Additional prospective research is needed to examine associations between psychosocial factors and diabetes risk in populations without any diagnosed diabetes. Studies should endeavour to control for all putative behavioural and socioeconomic covariates, such smoking, drinking, sleep disturbance, physical activity and treatment adherence. The measurement of possible psychobiological markers that might mediate the association between adverse psychosocial factors and risk of diabetes, such as neuroendocrine function (e.g. cortisol, catecholamines and neuropeptides) and immunological processes (diabetes-related autoantibodies) would also advance this field.

Conclusion

The current findings on the association of adverse psychosocial factors with poor diabetes prognosis point to the value of further research in the field, but they should be interpreted with caution and additional prospective research is needed.