The Socioeconomics and Quality of Life study simultaneously examined iSES factors and quality of life in adolescent and young adults with leukemia and lymphoma. Questionnaires were distributed to recently diagnosed leukemia and lymphoma cases recruited from the California Cancer Registry. Individual-level variables related to socioeconomic status and quality of life were collected and analyzed. Survivors recruited for the study were diagnosed with primary cancer (January 1, 2006–December 31, 2007) while between the ages of 15 and 40 and residing in the southern California—Orange, Imperial, and San Diego Counties.

Study procedures closely approximated protocol developed for the Kids, Adolescents, and Young Adults Cancer survey described elsewhere [14]. Data collection spanned from December, 2008 to December, 2009. Potential participants were sent a study packet that included: a cover letter describing the study and explaining how identification was accomplished through the cancer registry, the study questionnaire, a study information sheet that served as a waiver of written informed consent, a decline form to opt out, and a prepaid reply envelope. If a patient had a Spanish surname, as listed in the registry’s data dictionary, a certified bilingual (Spanish–English) letter was sent with the packet, followed by study materials in Spanish upon patient request. Project staff made follow-up phone calls to verify receipt of the study packet and answer any questions. Before any contact attempts, efforts were made to determine whether a patient had passed away, using tracing methods such as the Social Security Death Index and Alumni Finder (www.alumnifinder.com). All survivors who participated were sent a thank you letter and a $20 money order to compensate them for their time.

Population recruitment included 320 young adult cancer survivors (see Appendix 1, Fig. 1) for a flow diagram of the population-based recruitment). A total of 104 survivors from this recruitment with first-primary leukemia or lymphoma completed participation in the study. An additional six participants who fit eligibility criteria were recruited through a university-based cancer center to participate in an ancillary focus group study also completed the survey and were included to increase sample size. One of the population-based participants completed less than half of the survey with questionable data quality; therefore, these participant’s data were removed from further analyses. A total of 110 participants were included in the final analyses, which reflects 77 % of the participants who either participated or declined. However, only 43 % of the identified participants completed the survey. Quality control was performed by having the data recorders double-check each questionnaire entry and data completion with a sample (15 %) of questionnaires triple-checked by the first author.

Socioeconomic status was measured at two geographic scales, neighborhood-level (nSES) and individual-level (iSES), and conceptualized as a multidimensional construct following Galobardes et al.’ [15] definition: “the social and economic factors that influence which positions individual or groups will hold within the structure of a society”. The nSES measure included in the California Cancer Registry was originally created from census 2000 indicators for the California Teacher’s Study, an ongoing prospective cohort study aimed primarily at examining breast cancer etiology. Yost et al. [11] used principle components analysis to create a composite index of SES, selecting seven final indicators at the census-block group level for the measure: an educational attainment index, proportion with blue-collar job, proportion older than 16 in the workforce without a job, median household income, proportion below 200 % of the poverty level, median rent, and median house value, which all loaded as a single component, accounting for 59.7 % of the variance in the original data. This measure has been widely used in the analyses of health outcomes in California Cancer Registry data [16–23]. Quintiles of this measure were used in the current study [11].

Individual SES (iSES) was created in the current study using factor analysis (using SAS PROC FACTOR) from participants’ responses to iSES questions (see Appendix 2, Table 6). The number of factors retained was based principally on the prior assumption of a tridimensional structure of iSES from Oakes and Rossi’s identification of human-, material-, and social capital [2] and examination of scree plots, which indicated a parallelism between the second and third factor, further supporting our theoretical approach of keeping three factors. Oblique rotation was used to allow for correlation among factors [24], given the correlation of socioeconomic indicators like income, education, and social class. Items with loadings greater than 0.3 on at least one factor were retained. The analysis revealed one dominant factor which accounted for two-thirds of the common variance among the items. Indicators representing human capital (skills and knowledge) and material capital (tangible goods and economic resources) [2] had high factor loadings on this factor, which suggested a commonality between these two dimensions and was thus labeled ‘material and human capital’. Among young adults, particularly those suffering from a chronic disease, it may be that these dimensions are intertwined, as the cancer experience may impede educational goals and limit social mobility [25].

Indicators of social capital did emerge as representative of two distinct constructs: social support and social connections. Items including the ability to obtain a lawyer if needed, general trust in others, and reliance on family and neighbors all appear to indicate a measure of social support (Factor 2: Social Support). Income may be a more transient indicator of access to material resources, whereas wealth may reflect more stable access over time [15]. Net worth, which had a moderate loading on social capital, helps establish social strata and defines social boundaries, thus in part determining one’s level of social support, an important buffer during a stressful time [26].

Finally, Factor 3, described as social connections, had moderate-to-strong loadings on the number of professional friends, relationships with community leaders, and involvement in social groups. This factor seems to relate more to the size, rather than the quality, of the social network. Granovetter’s “strength of weak ties” theory [27] asserts the power of casual acquaintances over deeper bonds for transmitting information through a social network. In this way, having many social connections may provide more access to resources, and in terms of experiencing cancer, this could benefit an individual in terms of care-seeking and quality of life. Formal participation in social groups has also been found to be related to good health among adolescents [28].

The three inter-factor correlations (Appendix 3, Table 7) were weak to moderate, further justifying the three-factor solution. The overall Kaiser–Meyer–Olkin measure of sampling adequacy was 0.80 indicating a commonality among the variables selected, and the correlation matrix was singular and identifiable. The three-factor scores were summed with weights equal to unity to create a final composite iSES score for each participant [2] and quintiles of the score were generated for regression analyses. As expected, raw iSES scores were moderately correlated with the raw nSES score (r = 0.49). Individual SES items were all under 12 % missing except household income and net worth, which were 32 and 33 % missing, respectively. Missing values were imputed using the expectation maximization algorithm (using SAS version 9.2 PROC MI) with five imputations at each run [29], following a three-stage process. There was little evidence of bias due to the missing data. All further analyses of iSES were conducted on the final imputed dataset.

The Socioeconomics and Quality of Life survey included quality of life items developed by the investigators to address the specific domains central to the adolescent and young adult population. Previously published and validated quality of life scales did not appear to adequately address the specific concerns of adolescent and young adult cancer survivors, but rather young adult survivors of childhood cancers [30]. This is, perhaps, due to the dearth of quality of life research in this age group relative to younger and older counterparts [31]. As such, items were chosen to address age-specific issues such as identity, independence, relationships with family/friends, and the development of life skills. The four domains measured were physical health, psychological and emotional well-being, social relationships, and life skills. Several items from the Functional Assessment of Cancer Therapy-General [32] and the Youth Quality of Life Survey [33] were adapted to address global quality of life issues such as having pain or worrying about death. Respondents were asked to rate how much they agreed with each statement at the time of the survey using a Likert scale with possible responses of “not at all,” “a little,” “quite a bit,” “very much,” and “don’t know.” Negative items were reverse-coded in the analysis to create monotonic dimensions. Subscale scores of the four quality of life domains were examined. A draft of the survey was pretested in a mailed small pilot survey study of 15 AYA cancer survivors selected from the cancer registry. Pretest participants were asked to provide feedback on the survey items, which was incorporated into the final version of the survey. Cronbach’s alpha for the subscales were as follows: physical health = 0.71, psychological and emotional well-being = 0.70, social relationships = 0.65, and life skills = 0.67 and are consistent with levels deemed acceptable by a recent review on scales used in quality of life research [34].

Ordinary least-squares regression models were run on quality of life subscale mean scores. All analyses were conducted using SAS (version 9.2) PROC REG and PROC GLM functions. Preliminary analysis was conducted to ensure that assumptions for regression modeling were met. Regression diagnostics revealed no evidence of outliers (using Mahalanobis distances) or problems of multicollinearity (using variance inflation factors). Regression coefficients and adjusted r² values were compared from nested multiple regression models. Nested regression models were analyzed to examine the utility of iSES and nSES after other demographic covariates were entered to ensure that these measures of SES significantly contributed to the models. We were interested in comparing the relative contributions of both SES measures to determine whether collecting patient-level iSES data are worth the financial expense and administrative burden. All results were considered significant where P < 0.05.

In all four domains, the level of iSES was significantly associated with quality of life in the reduced and full models (P < 0.05). Yet, nSES was only significantly associated (P < 0.05) with physical health. The preliminary regression models revealed that domestic partner status (currently married/living with partner or otherwise), time since diagnosis, and receiving first-course chemotherapy or radiation treatment did not significantly predict quality of life; thus, they were not included in the final models.

The relative importance of both iSES and nSES on quality of life while accounting for demographic variables (age at diagnosis, gender, and race/ethnicity) was assessed through multiple linear regression using nested models. Models also adjusted for health insurance status at diagnosis, which is often more readily available and sometimes used as a proxy for SES [35], to control for potential confounding. Tables 2, 3, 4, 5 give unstandardized parameter estimates and adjusted r² values for models comparing the relative influence of SES variables on quality of life.

For each of the domains, the biggest improvement in model fit, as measured by changes in the adjusted r² values and log-likelihood ratio tests, occurred with the addition of the iSES score (changes in r²: physical health: +14 %; psychological and emotional well-being: +12 %; social relationships: +25 %; and life skills: +10 %). The addition of nSES to the reduced model did improve model fit, but only substantial changes were observed for physical health (physical health: +13 %; psychological and emotional well-being: +1 %; social relationships: +1 %; and life skills: −0.07 %). The multivariate nested model analysis suggests that iSES is more strongly related to QOL than nSES and that nSES likely does not provide much additional information on variation in QOL among young adults with leukemia and lymphoma, except perhaps for physical health.

The multidimensional approach used herein to operationalize iSES is a strength of the current study and responds to several recommendations about the measurement of SES in health research put forward by Braveman et al. [48]. These are the use of items appropriate for measuring the chosen study population; careful consideration of the mechanisms by which lower iSES negatively impacts quality of life; inclusion of a multitude of resource-relevant items in addition to the traditional measures of education, income, and occupation; and specification of separate dimensions of SES along with comparison to single indicators.

The iSES variable under analysis was derived with exploratory factor analysis, and more rigorous psychometric testing should be undertaken to ensure it is a robust measure. The small sample size limited our statistical power overall and the ability to conduct analyses stratified by age groups, gender, or race/ethnicity. Only leukemia and lymphoma survivors were included in the analysis, and further research should include survivors of other cancer types as well as make comparisons to healthy control populations. The quality of life measure developed for this study had weak-to-modest inter-item correlations and was not validated due to budgetary and time constrictions. While the current study made use of items from two well-validated scales, future research would benefit from a quality of life scale and is validated for adolescent and young adult cancer populations [34]. Longitudinal studies measuring change in quality of life as well as change in relevant SES predictors would provide more complete information for guiding clinical intervention efforts.

Participation rates in the current study appeared to be reasonable compared with other cancer survivor studies [49–52]. Utilizing bilingual recruitment materials helped to decrease non-participation rates among Hispanics in a population-based study of non-Hodgkin’s lymphoma survivors [53]. Because some information is available through the California Cancer Registry for all potential participants, demographic variables were compared for study responders and non-responders. More females (49 %) responded than males (36 %), but no other significant differences in terms of stage at diagnosis (lymphoma only), time since diagnosis, race/ethnicity, gender, age, or neighborhood SES in the current study were found. However, if individuals from high iSES backgrounds and better health were more likely to participate, it is possible we underestimated the impact of iSES on quality of life.

Cancer registries are a vital source of data for conducting population-based cancer studies, particularly in age groups with relatively low incidence or in rare tumor types. Populations may be better served if cancer registries were to include indicators of individual-level SES. The California Cancer Registry’s census-derived nSES measure is more precise than what is available in most cancer registries, which tend to measure SES at the county level. Registries may widely underestimate the impact of SES on health outcomes if area-level measurements are the sole indicator of SES.

Although there was no comparison group in this analysis, the relatively high mean domain scores suggest that, overall, young adult leukemia and lymphoma survivors do experience positive quality of life. The failure of many other quality of life studies to account for variations in social and economic factors eschews the importance of identifying health disparities and ignores the accumulated embodiment of socioeconomic stress [54]. The approach undertaken in this analysis assumes direct effects of SES on quality of life in this population, although more complex models with mediating pathways conducted on larger samples could further elucidate key potential points for intervention. Intervention studies designed to target SES poor survivors, perhaps with longer and more regular follow-up, should be developed. However, addressing these immediate needs should not supplant recognition of the inequitable impact of chronic health conditions on individuals with weak or nonexisting socioeconomic buffers.