Impact of early psychosocial factors (childhood socioeconomic factors and adversities) on future risk of type 2 diabetes, metabolic disturbances and obesity: a systematic review

We searched Medline in July 2008 and carried out an update of the search in April 2010 applying the search algorithm "(diabetes or insulin resistance or prediabetes or metabolic syndrome) and (SES or social or socioeconomic or psychosocial or income or working status or migration or community or adversities or deprivation or depression or abuse or high risk family or hostility)".

For obesity a search in Medline was carried out in November 2008 (updated in April 2010) using the algorithm" (obesity or overweight) and (socioeconomic or social or deprivation or adversities or childhood socioeconomic or family environment or early life or youth or childhood adversities or deprivation) and (longitudinal or prospective or cohort)".

Both searches were limited to publications after 31 December 1994 in English language. During the 1990 s, the diabetes criteria were changed several times, which complicates comparison with earlier publications.

Studies were included if incident cases of type 2 diabetes were assessed. Cross-sectional studies with data on diabetes prevalence only were excluded. Only population- or community-based studies were included to have a more homogeneous database for this review. Further outcome criteria were insulin resistance, elevated HbA1c values, and the metabolic syndrome because they share risk factors and pathophysiological pathways with type 2 diabetes. Obesity and overweight are considered key risk factors for type 2 diabetes incidence especially in youth, and were furthermore included as outcome parameters, but were analysed separately.

Studies on incident type 2 diabetes or obesity were included if they contained data on psychosocial factors including a wide range of indicators reflecting the social and psychological conditions under which the participants grew up:

(1) basic indicators of CSES: parental education; parental occupation and family income as indicators of economic wealth and stability;

(2) further indicators of wealth and deprivation: public housing, housing conditions, house ownership, unemployment;

(3) the high-risk family concept (adverse childhood experiences) regarding neglect, abuse and household dysfunction [24];

(4) indicators of impaired psychological health in children such as depression or anxiety;

(5) indicators of the ability to cope with stressful conditions like coping skills or sense of control [25];

(6) indicators of possibly stressful situations: migration, parental stress (e.g. stressful working conditions or parenting stress which are supposed to have implications for children's stress as well);

(7) neighbourhood deprivation indexes as indicator for the aggregation of unfavourable circumstances clustered in residential areas. These indexes are based on the social composition of neighbourhoods regarding for example the social status of its habitants, housing and street conditions, mean household sizes per person and other indicators of wealth such as mean household number of cars [26, 27].

Studies were excluded if information on obesity and overweight was only available at one time point and no adjustment for previous weight status was done. Studies, which only included type 1 diabetes were also excluded. Self-reported diabetes type may not offer a reliable distinction between the diabetes mellitus types. However, self-reported diabetes type was found in most studies and as type 1 diabetes contributes to a fairly small proportion of the total number of diabetes cases we accepted self-reported diabetes type for the assessment of type 2 diabetes incidence.

Eligible studies were assessed by one reviewer (T. T.) and discussed with a second reviewer (W. R.). The final list of variables extracted from the selected studies contained the first author, publication year, country where the study was carried out, cohort size, study duration, age characteristics of participants, measurement of risk factors, definition of diabetes/obesity variables and a brief description of results including effect size. For missing information we contacted some of the authors of the selected studies. Any disagreements regarding numbers, study inclusion, and further analysis were resolved by consensus between the authors (T.T., C.H., W.R.).

We restricted the analysis to descriptive measures because of the lack of statistical comparability for most studies. Effect sizes extracted from the publications reflect adjusted results for odds ratio (OR), hazard ratio (HR) and relative risk (RR) regarding confounding (1) for age, sex, BMI, physical activity, smoking and alcohol (for type 2 diabetes incidence) or (2) for birth weight and adult SES (for obesity and overweight). We developed a 5 item quality scale adapted to our purpose with reference to methodological recommendations of the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) group [28, 29]. The sum score derived from this scale mainly had to reflect the study design, the examined outcome parameters, the risk factor construction and the control for confounders. Quality was judged taking into account the current state of research on type 2 diabetes and obesity and possible sources of bias. The following criteria were included into the quality score for the (1) diabetes and (2) obesity studies:

• (1&2) Study Duration (D): Highest quality was considered to be obtained from birth cohorts (D = 1) as the future diabetes and obesity risk is influenced strongly by very early health indicators such as birth weight and gestational age. Furthermore, especially in childhood body composition and metabolic functioning vary in different age groups, so that homogeneous age groups as accomplished by birth cohort studies offer another quality advantage [30, 31].

• (1&2) Recruitment (R): Our inclusion criteria comprised only studies from which a high grade of representativeness and reproducibility was expected. To further assess the quality of recruitment in the score, population-based or school-based studies with characteristics of a census were rated as of highest quality (R = 1).

• (1&2) Explanatory variables/risk factor specification (E): We expected a source of bias from retrospectively assessed childhood psychosocial factors in the offspring (E = 0). In case parents gave direct information on their occupational, educational and financial situation we rated the quality of risk factor construction as high (E = 1).

• (1) Outcome parameter diabetes: Highest evidence was expected from blood glucose measurements with internationally valid cut off-levels as defined by the American Diabetes Association or WHO (O = 1) [32]. Self-reported diabetes leads to an underestimation of type 2 diabetes cases due to a high number of undiagnosed cases [33, 34]. Therefore, we considered the quality based on self-report as low (O = 0). The plasma-based measure of insulin resistance (HOMA-IR calculated from fasting glucose and insulin levels) is closely related to (pre)diabetes. Thus, evidence from HOMA-IR-based data were considered as high (O = 1). Although HbA1c has been suggested as diagnostic tool for type 2 diabetes (ADA 2010), especially in early diabetes and in prediabetes, HbA1c values lead to misinterpretation which may be also due to a genetic component in the metabolisation of glycated haemoglobin [35‐39]. Thus, we rated the quality of HbA1c values as low (O = 0). Metabolic syndrome and type 2 diabetes share some risk factors, but are too different as entities to include the metabolic syndrome as surrogate for type 2 diabetes. We allowed for the term 'metabolic' during Medline search, but rated metabolic syndrome as outcome criteria of low evidence for our study question (O = 0).

• (2) Outcome parameter obesity (O): First, measured weight and height to calculate BMI met our quality demands. Self-reported weight and height on the other hand have been reported to be imprecise in adults and in parents reporting anthropometric data of their children. Standard cut-off values for overweight and obesity served as second indicator for high quality: In children overweight and obesity are defined based on the age and sex-specific 85^th and 95^th BMI percentiles in growth charts from the National Health and Nutrition Examination Survey (NHANES), the Centers for Disease Control and Prevention (CDC) or the International Obesity Taskforce (IOTF), whereas in adults cut-off values for BMI ≥ 25 kg/m² for overweight and ≥ 30 kg/m² for obesity were used. The highest score (O = 1) was given only if both criteria were fulfilled.

• (1&2) Confounding (C): Quality of adjustment for confounders was inferred from important behavioural pathways leading to type 2 diabetes. Adjustment of age, sex, BMI, smoking, physical activity and alcohol consumption was required for the highest quality score (C = 1). Adjustment for obesity in the extracted studies was heterogeneous and depended highly on the study question. According to the basic requirements for our review question a high study quality (O = 1) demanded at least control for birth weight or baseline BMI to have possible weight change considered and the adjustment for adult SES to control SES change throughout life.

The search strategy yielded a total of 19,504 results. After exclusion of 5,537 non-English articles, reviews, animal studies, case reports and articles unrelated to type 2 diabetes, 13,967 abstracts were screened (see also Figure 1). We assumed that social parameters are often treated as confounders and in this case the results are presented in full text articles, but not in abstracts. Therefore, the first screening of abstracts served mainly to identify articles on diabetes incidence (591). Of these, 280 did not meet the study design criteria and were excluded. The further thematic evaluation of the psychosocial relevance of the studies was based on full text articles. Of these 273 were thematically irrelevant mainly because they did not offer information on childhood psychosocial factors. One study retrieved during the update of the search was not added, because it presented race stratified results and there was already non-stratified data available from this study from the first search [40].

For reasons of limited statistical comparability (p.eg. limited presentation of data) 10 studies remained for further descriptive analysis (Table 1). Information on excluded studies with a unique method of analysis such as path models is provided in the discussion, information on excluded studies with solely examined risk factors in the results.

The second search strategy for the outcome obesity retrieved 1,631 results (see also Figure 1). 77 reviews and non-English publications were excluded. 1,428 studies were classified as thematically irrelevant and 76 publications did not meet study design criteria. This led to 50 longitudinal studies of which 36 were not comparable for methodological reasons and risk factor selection. Hence, 14 publications (13 studies) were included for this review and are presented in Table 2.

In four studies, the risk of developing type 2 diabetes or metabolic disturbances in the offspring of lower social classes compared to children of higher status was only slightly elevated after adjustment with different effect measures (OR, HR and RR) ranging between 1.08 and 1.7 [41, 42, 46, 48]. In the Alameda County Study, disparities in self-reported diabetes were most striking in overweight women with low childhood socioeconomic position [42]: The odds of developing type 2 diabetes was 3.2 fold higher for them than for their high SEP counterparts (OR: 2.9 (95% CI 1.7; 4.8) vs. 0.9 (95% CI 0.4; 2.2)).

As shown in Table 1, the study with the smallest population size (N = 233) showed the largest protective effect (men OR 0.2 (95% CI 0.05, 0.8); women 0.6 (0.1, 2.7)) of a high social status (father's occupation) [49]. This study was based on middle-aged participants of the Newcastle Thousand Families Cohort with a BMI exceeding 25 kg/m². Furthermore a large protective effect of maternal education (beyond elementary level) was found in a Mexican population (OR 0.6 (95% CI 0.5; 0,8)) [44]. In contrary, low maternal education at high school level or less was a significant factor influencing type 2 diabetes incidence in the Princeton School District Study (ß-coefficient = 4.47 (SE 0.78)).

On the other hand, two studies indicated small, non-significant protective effects of low social class. In a Finnish cohort children at the age of 7 years from blue-collar families had a reduced risk (OR 0.8 (95% CI 0.48; 1.45) of being diagnosed with type 2 diabetes in comparison with their white-collar counterparts [43]. In a British Cohort only women of the lowest classes were less likely to have HbA1c levels exceeding 5.8% (OR 0.8 (95% CI 0.5; 1.4)). In men, no effect of social class was seen in this study (OR 1.1 (95% CI 1.0; 1.8)) [47].

The effect sizes observed in the included studies (except for beta coefficients) are shown in Figure 2. Effects of high social status have been inverted for better visual comparability.

Furthermore, we were interested in the interaction of childhood and adult psychosocial variables. SES change was explicitly examined in the Nurses Health Study [41]. Participants in this study were relatively homogeneous with respect to their educational level and occupational status as all 100,330 participants were female nurses. Disparities emerged from different childhood socioeconomic positions and from husband's educational level. Improving SES over lifetime (spouse's high educational level and low father´s occupational status) with stable high SES as reference resulted in a slightly reduced, albeit non-significant relative risk for type 2 diabetes (RR 0.9 (95% CI 0.7; 1.3), whereas a stable intermediate SES and a declining SES influenced the type 2 diabetes relative risk negatively (RR 1.2 (95% CI 1.06; 1.4) and RR 1.18 (95% CI 1.06; 1.3), respectively). The relative risk in participants with stable low SES was comparable to those participants with stable high SES.

Langenberg et al. examined lifetime effects by analysing the change in the influence of the socioeconomic position on HbA1c levels with adjustment for childhood and adult SEP [47]. In both men and women, adult social class had stronger effects than childhood social class on HbA1c values exceeding 5.8%. In women childhood effects on HbA1c even reversed in the fully adjusted model. Langenberg et al. interpret these findings by postulating that low childhood social class continues to influence adult social class as both variables were highly correlated [47].

A high continuity of childhood and adult SEP parameters was also observed in a Finnish cohort, where children from manual classes were more likely to work in manual or lower non-manual occupations in adulthood [46].

Among the retrieved studies there were several unique findings that were not comparable with other studies and other results. However, most of these studies revealed a possible association with future diabetes.

For example psychological factors such as childhood adversities [48] were associated with elevated HbA1c and obesity in one study. After full adjustment, these associations were no longer significant for elevated HbA1c. A further possible psychological risk factors for diabetes among our findings were depression and anxiety [70, 71], hostility [72], and sense of coherence [73].

Also, indirect measures of socioeconomic status such as deprived neighbourhoods, and housing conditions [27, 28, 74, 75] are likely to be relevant for future diabetes incidence. Neighbourhood characteristics also have influence on lifestyle habits depending e.g. on the availability of healthy foods and on the number of facilities for physical activity [76].

Overall, more comparable studies are needed to quantify the association of other childhood factors on metabolic impairment.

Table 2 gives an overview of the included studies and the observed effect sizes which are furthermore visualised in Figure 3.

Effects were most prominent in two studies regarding income discrepancies: In a national longitudinal survey in the USA the RR in the lowest intracohortal ( < 15^th percentile) income group for obesity incidence was 2.84 (95% CI 1.39; 5.78) compared to the highest ( > 85^th percentile) income group during 6 years after the baseline examination [61]. Similar results were observed in a Canadian birth cohort at the follow-up of 4.5 years [58]. At that time point the lowest income group (less than 20,000 CN$/year) had a 2.5 fold increased odds (95% CI 1.3, 4.8) of being overweight at the age of 4.5 years with reference to families with an annual income of 60,000 CN$ or more. These results were obtained after adjustment for gestational age and birth weight. One study observed that belonging to a low income family (less than 26,000 US$/year) was a significant predictor of being obese at an earlier age (β-coefficient: -2.6 (95% CI -3.8; -1.3)) [62]. No effect of family income was seen in another study which included data from the National Longitudinal Surveys Child-Mother files in the USA (OR 1.0 (95% CI 0.99, 1.00)) [57].

Parental education had no direct or a small influence on overweight or obesity outcomes in six studies [57, 58, 60, 61, 68, 69]. In the National Longitudinal Survey of Youth [61], the Home Observation for Measurement of the Environment-Short Form (HOME-SF) accounted for a modulation of other CSES factors particularly for maternal education. The unadjusted OR for the six-year cumulative incidence of childhood obesity for maternal education lower than high school was 1.47 (95% CI 1.04; 2.1). The fully adjusted OR was lowered to 0.96 (95% CI 0.7; 1.4) when including also HOME-SF as covariable. In contrast, family income seemed to be independent of the HOME-SF in this study [61].

In one study regarding parental occupation gender disparities were prominent [67]. Giskes et al. found a significantly higher risk for baseline overweight and obesity (at 40-60 years) for female participants whose fathers had been working in blue-collar occupations (OR 3.4 (95% CI 1.9; 6.1), father's professional occupation as reference). This effect decreased slightly after adjustment for adult SES (OR 2.8 (95% CI 1.6; 5.2)). These women also gained significantly more weight between baseline and follow-up 13 years later (OR 2.0 (95% CI 1.7; 2.3)). In another study the risk for developing overweight during the 6-year follow-up period was increased for children from blue-collar families (OR 2.4 (95% CI 1.02; 5.4) in comparison with their white-collar counterparts [65]. Furthermore, in two studies the adjusted effects of low status were small [54, 55].

Based on results of regression analysis, two studies determined low social status as factor influencing overweight and obesity [56, 62], whereas in a third study no such effect was seen (ß-coefficient: -0.05 (SE 0.08)) [66]. In another study, DXA-assessed fat mass was analysed using a slope index of inequality (SII). The results of this method are comparable to regression coefficients and pointed towards a moderate association of maternal education on body fat [59].

However, as these studies only displayed data on regression coefficients, effect sizes cannot be compared to the other studies and are therefore not included in Figure 3.

Taken together, whenever effects were seen in any of the thirteen studies, they pointed towards a deleterious influence of low social status on future risk of overweight and obesity.

The role of change in psychosocial factors in relation to obesity was investigated in the 1958 British birth cohort [55]: Power et al. reported a continuous decrease of the effects of low socioeconomic status in childhood on obesity especially in women with an increase in social position in adulthood. The effect of low social class at the age of seven years in 33-year old women decreased from an unadjusted OR of 1.4 (95% CI 1.3; 1.6) to 1.3 (95% CI 1.1; 1.4) after adjustment for personal education [55].

We found one study which showed that traumata are associated with obesity [77]. Furthermore, depression and stress [78] are not only relevant for type 2 diabetes, but also for obesity. Also, indirect measures of socioeconomic status such as deprived neighbourhoods, housing conditions [79] and supply for healthy lifestyle habits [80], are likely to be relevant for obesity.

Therefore, psychological factors, and several measures of deprivation in childhood remain an interesting field for further examinations helping to understand the possible pathways leading to obesity and type 2 diabetes.

Altogether we observe that for both type 2 diabetes and obesity longitudinal, population-based data are scarce. Especially, the life course including childhood indicators of psychosocial status and the role of risk factor clustering is under-investigated. Additionally, analysis of psychosocial factors is carried out heterogeneously and thus exposed to the critique of constructing SES risk factors arbitrarily. Although there are some national consensus statements especially regarding the assessment of the occupational position [51‐53], it is difficult to compare these positions because they are attributed with varying amounts of prestige and are influenced by general shifts in composition [81].

Three steps may help to gain further evidence on the topic: First, basic SES measures in youth and in adulthood should be consequently presented in publications. Second, for a better understanding of the role of psychosocial factors throughout life on the risk of type 2 diabetes, detailed analysis of a broad set of psychosocial factors, their interrelation and their impact on type 2 diabetes are needed. This requires a large-scale systematic analysis applying various psychosocial measures in youth and adulthood that have been found to be of relevance for diabetes and obesity such as depression [82‐84], stress [85, 86], unemployment [87], lifestyle habits [76, 80] deprived neighbourhoods [27, 28, 74, 75, 79] and other factors [72, 77].

Finally, such a large-scale systematic analysis would also require the application of different analytic approaches. As an example, the interpretation of the results of regression models analysing closely related risk factors throughout lifetime is controversial. Attenuation after adjustment for various risk factors may be attributable to a strong correlation of these risk factors [22] rather than to confounding. An interesting alternative approach may be the analysis of path models. For example Lehman et al. showed that childhood SES had a direct impact on metabolic functioning in the participants by applying path models on data from the Coronary Artery Risk Development in Young Adults study (CARDIA). Furthermore, childhood SES had an impact on early family environment, psychosocial functioning and adult SES. Interestingly, adult SES in this study had no direct influence on metabolic functioning [88]. Additionally, cluster and discriminance analysis can shed further light on the interaction of a broad set of childhood and adult psychosocial indicators accounting also for highly differentiated social milieus with distinct beliefs, behaviours and tastes [89].

Valid results on the association of childhood socio-economic circumstances and future risk of diabetes and obesity would be important to design targeted and more efficient prevention strategies. Diabetes and obesity prevention may not only profit from educational programmes but also from health politics, from interventions for high-risk families, from coping skills training [90], from empowerment of social networks and from healthy neighbourhoods [91, 92].