Life course socio-economic position and quality of life in adulthood: a systematic review of life course models

Search strategy

Articles were identified by searching (via Ovid) the electronic databases Medline (1948-present), Embase (1947-present) and PsycInfo (1987-present). Additional searches were executed in Web of Science and Cambridge Scientific Abstracts (CSA) Illumina. Web of Science covered the databases Science Citation Index Expanded (1945-present), Social Sciences Citation Index (1956-present), Arts & Humanities Citation Index (1975-present), Conference Proceedings Citation Index- Science (1990-present) and Conference Proceedings Citation Index- Social Science & Humanities (1990-present). CSA Illumina covered Applied Social Sciences Index and Abstracts (1987-present), International Bibliography of the Social Sciences (1951-present), CSA Sociological Abstracts (1952-present), and Worldwide Political Science Abstracts (1975-present). All searches were carried out on January 2^nd 2012 and limited to English language articles. No restrictions were placed on the publication date of articles. Reference lists of included articles were checked for any additional articles and citations were accessed via Google Scholar and checked manually.

Searches included terms used to describe SEP, such as ‘social class’ and ‘occupation’, combined with terms used to describe the life course and quality of life. Relevant MeSH headings were used when available. See Additional file 1 for a full example of the search strategy executed in Medline.

Eligibility criteria

Studies were included if they met the following criteria: primary studies published in a scholarly journal; based on populations within industrialized countries as defined by Organization for Economic Co-operation and Development criteria [27]; reported subjective quality of life as an outcome (using indicators separate from physical health such as wellbeing, life satisfaction or specific quality of life scales such as CASP-19); reported outcomes in males and/or females aged 25 years or over (as this represented an adult population in which individuals were likely to have completed their education); contained individual or household measures of SEP from at least two time points (childhood and adulthood, or two time points in adulthood, regardless of the length of time between measurement points or whether they specified a particular life course model).

Articles were excluded if they contained only qualitative data, were review articles, did not specify any information regarding the age of participants, included only area-level or subjective measures of social status (as these were considered different constructs [28]), only looked at employment status, job or income mobility (without a measure of social class), contained only measures of physical health-related quality of life or which did not separate between physical and mental components of health-related quality of life (as we were interested in outcomes capturing aspects of quality of life separate from physical health), included only individuals with specific health conditions (e.g. dementia, psychiatric illness) as their population of interest, or contained only outcomes relating to psychiatric symptoms (e.g. psychological distress or depressive symptoms).

Study selection and data extraction

Title and abstract screening for immediately irrelevant articles was performed by one reviewer (CLN). Two reviewers (CLN and SVK) independently assessed the full-texts of articles short-listed against the eligibility criteria. Disagreements were resolved by consensus. All records were stored in Endnote X4. An Excel proforma was developed to assist in the data extraction procedure and included: the publication information (authors, year, journal), study characteristics (sample size, study design, response and attrition rates, time period), participant demographics (age at recruitment, gender, country), measurement of SEP (collection method, age at measurement, missing data), outcomes (summary measure such as mean quality of life scores or odds ratio of experiencing low quality of life, collection method, age at measurement, missing data), analysis methods (statistical techniques, variables controlled for, treatment of missing data) and results. Data were extracted by CLN and checked by SVK.

Quality appraisal

Quality appraisal was performed using an adapted version of the ‘Quality Assessment Tool for Quantitative Studies’ [29]. The following items were used to assess the quality and risk of bias within studies: sampling method, sample representativeness, study design, response rates, attrition rates and reasons (including death and loss to follow-up), whether the characteristics of those lost to attrition or non-response differed from those of responders, measurement of SEP and quality of life variables, reporting of missing data, and variables controlled for in the analysis to reduce confounding. Given the limitations associated with scoring criteria [30], the quality of articles was initially considered during the synthesis process. Three key items, which we considered the most important quality criteria, were then selected to provide an overall indication of study quality using a rating system. These were: the response and attrition rates, measurement of SEP, and sample size. A grade of higher, average or lower quality was given to studies based on the sum of scores for these items (see Additional file 2 for full details of the rating system). The quality appraisal of studies did not differ between the two methods of assessment (synthesis based on all quality items extracted and the rating system using the three key items).

Data analysis and presentation

Studies differed in terms of the measures of SEP, outcomes, time points considered and analysis techniques implemented. For these reasons, meta-analysis was not appropriate and narrative synthesis was used. Studies were categorized based on the life course model analyzed: accumulation, latent, pathway, or social mobility (inter-generational and/or intra-generational). Studies were grouped into the relevant life course model based on their aim, analytic approach and findings, similar to the method by Pollitt et al. [3]. We also compared our own classifications with those of the authors, if provided, but no conflicting groupings occurred. If more than one model was assessed within the same study, the results are presented under multiple categories according to the life course models investigated. It should be noted that positive results may be found for more than one model within the same study. A summary of the evidence is presented under each life course model (studies considered higher quality are described first), followed by a discussion of the key issues relating to the comparison of results.

Search results

The electronic searches identified 7,529 publications, following removal of duplicates (Figure 1). Thirty-seven additional records were identified via reference and citation checks. Following title and abstract screening, 7,486 irrelevant records were excluded. Examples of records excluded at this stage included those assessing health-related quality of life in disease-specific patient groups. Eighty full-text articles were assessed for eligibility, of which, 12 were selected for inclusion. Eleven articles were included by CLN; an additional article was identified by SVK and included following discussion. See Additional file 3 for the list of excluded full-text articles and the corresponding reasons for exclusion.

Study characteristics

Quality assessment summary

The sample size of included studies ranged from 91 [37] to 8,970 individuals [31, 32], with a median of 2,117 individuals. The attrition rate of studies ranged from 3% over 14 years [6], to 57% over 27 years [37]. Response rates for the last wave of study reported varied from 61% [38] to 90% [39], median 76.5%. See Additional file 5 and Additional file 6 for details of the full quality appraisal and ratings for each article.

Accumulation model

Three studies tested the relationship between life course SEP and quality of life using the accumulation model [6, 32, 40]. All three studies were assessed as being of average quality. Singh-Manoux et al.(2004) found some support for the model among men; as the number of occasions in a low SEP increased, the more likely the respondent was to report MCS scores in the lowest category [6]. A significant linear trend was also identified. However, no evidence was found among women, with the highest odds ratio being found in those who had an intermediate SEP score. Highest odds ratios for men were evident in trajectories that began in an intermediate or low SEP during childhood and moved to a high SEP in adulthood, perhaps indicating support for latent or social mobility models. Otero-Rodriguez et al. (2011) used polytomous logistic regression models with risk of decline, no change (reference category) and improvement in MCS scores over a two year period as the outcome. They found the risk of decline was highest in individuals reporting three low SEPs at key points in the life course, compared to those with one low SEP, and a significant linear trend was identified [33]. The risk of improvement was also highest in those reporting three low SEPs, but no linear trend was identified. Mäkinen et al. (2006) found no support for the accumulation model using education level as the only indicator of SEP [32]. The prevalence of MCS scores with ‘limited functioning’ was slightly increased for men and women who experienced a low childhood SEP and high adulthood SEP, compared to those with a low SEP at both time points.

In summary, mixed evidence was found for the accumulation model and differing results were found between genders. The contradictory results of Mäkinen et al. (2006) could be explained by heterogeneity between studies in terms of the SEP measures and study contexts. Mäkinen et al. (2006) used the respondent’s own education level and parental education level, whereas Singh-Manoux et al. (2004) used a combination of the respondent’s employment grade, education level, and parental occupation. Perhaps education levels alone do not capture accumulated socio-economic disadvantage to the same extent as taking parental and respondent’s occupations into account. Education level may also be considered a more distal indicator of SEP, which reflects both prior cognitive ability and parental SEP. Cohort differences in the meaning of education between generations may also exist. For example, Mäkinen et al. (2006) used a Finnish population aged 40 to 60 years old. It is likely that the respondent’s parents completed their education at a time before the rapid expansion in higher education. Over half of respondents had high education levels, but over half of their parents had low education levels [32]. Therefore, having low education level in the parental generation may not confer disadvantage in the same way as in the respondent’s generation, where a higher education level may confer higher social status. The context of the paper by Singh-Manoux et al. (2004) is also important to consider. They used the Whitehall II study based on British civil servants, in which the social hierarchy is more amenable to measurement due to the distinct civil service grades.

Latent and pathway models

Five studies assessed life course effects on quality of life using the latent model [31, 32, 38, 40, 41]. Laaksonen et al. (2007) and Mäkinen et al. (2006) used the same dataset, but different methods, to simultaneously test the pathway model [31, 32]. One study was assessed as being of higher quality [41] and four were of average quality [31, 32, 38, 40].

Huurre et al. (2003) found poorer mean wellbeing scores among female respondents reporting their parents had a manual occupation, compared to non-manual occupation [41]. When adjusting for the respondent’s own current social class, this difference remained significant. The effect was not observed among men. Otero-Rodriguez et al. (2011) found that individuals reporting a low childhood SEP were more likely to experience change in MCS scores compared to those with a high childhood SEP [40]. Low education was also independently associated with a decline in MCS scores, but not improvement. Adulthood SEP did not have an independent effect on change in MCS scores. Marmot et al. (1998) found no support for the latent model among men and some support among women [38]. Using father’s education, the odds ratios for low wellbeing among women were not significantly different between education levels. However, using mother’s education level, the odds ratio for low wellbeing was significantly elevated among women with the lowest educated mothers. This indicates that, for women, a latent effect of low childhood SEP may operate via the mother’s SEP.

Conflicting results were found between the two studies by Laaksonen et al. (2007) and Mäkinen et al. (2006) [25, 26]. Mäkinen et al. (2006) used a similar method to the above studies to test the latent model and identified support among women, but in the opposite direction to that expected [32]. Women who had a low or intermediate SEP in childhood exhibited lower odds of having MCS scores in the ‘low functioning’ range, compared to those recording a high childhood SEP. This effect was not present in men. No significant pathway effect was suggested in men or women. On the other hand, Laaksonen et al. (2007) implemented a structural equation modeling approach and found no evidence for a direct effect of childhood SEP on MCS means [31]. However, they identified a direct effect of adulthood SEP in the opposite direction hypothesized; higher adulthood SEP was associated with poorer MCS scores in both men and women. The results indicated evidence for a pathway effect in which childhood SEP influenced MCS scores, via adulthood SEP which acted as a mediator. However, the authors highlighted that their method of analysis did not permit the calculation of confidence intervals around the indirect effects estimates and cautioned that this result should be interpreted with care.

To summarize, mixed evidence was found for a latent effect, with an indication that the model may be more plausible among women. Care should be taken when making overall conclusions due to differing exposure and outcome variables and inconsistent analysis methods. The results by Mäkinen et al. (2006) and Laaksonen et al. (2007) highlight that contrasting results may be found using different analysis techniques and different measures of SEP, despite using the same data. Modeling SEP as a latent variable including education, occupation and income, as in the paper by Laaksonen et al., may provide a more accurate indication of adulthood SEP and act to diminish any effect of childhood SEP. However, this does not explain why Mäkinen et al. (2006) found an association in the opposite direction to that expected. The two studies used data from a Finnish public sector occupational cohort where 80% of participants were female. Compared to the studies by Huurre et al. (2003) and Marmot et al. (1998), the women employed may work in jobs requiring higher demand, with higher status and stress. This could explain why Laaksonen et al. (2007) identified evidence for a pathway effect in which current, rather than childhood socio-economic circumstances had most influence on quality of life.

Social mobility models

Seven publications investigated the effect of social mobility on quality of life. Inter-generational mobility was tested in three [37, 39, 40], intra-generational mobility in five [33–35, 39, 42] and one assessed both types [39]. The quality of studies varied considerably. First, the results for inter-generational mobility will be presented, followed by those for intra-generational mobility.