Adverse childhood experiences and health-related quality of life in adulthood: revelations from a community needs assessment

Context

The project “Toward Eliminating Disparities in Maternal and Child Health Populations” is a 3-year CBPR (community-based participatory research) initiative that partners
researchers at the University of South Florida (USF), a non-profit advocacy and empowerment
organization, REACHUP, Inc., and a group of community representatives from selected
zip codes in Central Tampa, Hillsborough County, Florida. The project has three sequential
phases: 1) needs assessment to identify priority maternal and child health issues,
2) planning an evidence-based intervention that addresses the priorities identified,
and 3) implementation of a pilot community-driven intervention. The project is funded
by the National Institute on Minority Health and Health Disparities (5R24MD8056-02).

This paper focuses exclusively on the needs assessment survey, which was designed
to explore health determinants and quality of life indicators in the target community,
implemented using a CBPR approach. This CBPR study builds extensively upon a strong,
existing community-academic partnership. At the study’s outset, a Community Advisory
Board (CAB) was created, which consisted of eight adult residents from the target
community who were recruited based upon their knowledge, participation, advocacy,
and leadership in previous community projects 16], 17]. In addition to the required Human Subjects Protection Certification course, all
CAB members completed skill-building workshops in research methods. Active participation
of community members occurred in the design, data collection, and analysis phases
of the study.

Design

The community needs assessment survey was a cross-sectional instrument designed to assess factors associated with HRQoL
and was administered between November 2013 and March 2014. Two-hundred and one adult
participants were recruited from approximately 110,451 residents of the target population
18]. The target community was predominantly African American (60 % black, 18.3 % white,
12.1 % Hispanic, and 9.6 % other) and tended to be economically disadvantaged, with
half the income and double the unemployment rate of the rest of the county 19], 20]. The survey was administered through intercept interviews across a five zip code
area 21], 22]. Intercept interviews are a type of social marketing research in which respondents are approached by culturally
and linguistically matched research staff and “intercepted” in various places where
community members gather often 23]. This method is used to collect more representative information than convenience
samples regarding particular issues of concern to the community, and intercept interviews
are more feasible and less expensive than door-to-door needs assessments. Intercept
interviews are also useful when the target population is widely dispersed and harder
to reach (e.g., several zip code-level areas with economically disadvantaged urban
communities). Accordingly, trained CAB members surveyed people “in the streets” at
frequently used community locations (e.g., cafes, churches, libraries, local schools
gatherings, shopping centers) and at times when community residents were accessible
(e.g., weekends, after hours during weekdays). In a similar manner to 30 by 7 cluster
sampling 24], the CAB nominated 30 community locations or “clusters” (23 were actually surveyed
due to redundancy of participants in 7 locations) based on socio-demographic and economic
characteristics using zip code level census data, and then randomly sampled at least
seven individuals from each location. To participate in the survey, a respondent must
have been a community resident for at least the previous 12 months. Flyers, social
media, and “word-of-mouth” transmission of information were used for recruitment.
Written informed consent was obtained and a modest but appropriate monetary incentive
was given to study participants. Approval for the study was obtained from the Institutional
Review Board of the University of South Florida.

Measures

The development of survey questions was guided by the Life Course Perspective (LCP)
with extensive input from the CAB. The LCP framework is used to evaluate the cumulative
influences of risk and protective factors during critical periods of human development
and the effects those events have on the health trajectories of individuals 25]–27]. First, community members, in partnership with academic researchers, reviewed local
data and voiced their concerns for critical topics. Next, academic researchers suggested
questionnaires or scales that had been validated in previous research studies for
local adaptation. Subsequently, the CAB provided feedback on wording and readability,
question addition/deletion, assessed acceptability of questions and technology usability
(see Technology section), and pilot tested questions before community-wide implementation.
The final survey questionnaire contained 63 questions (Likert-type and multiple choice
questions), which covered the following inquiry domains and specific sequencing into
the survey: life in the neighborhood (neighborhood assets; community-wide issues),
social connections (social support), health and quality of life (general health, HRQoL,
self-reported health problems, sleepless days, perceived social standing), stress
and unfair treatment (stress appraisal, perceived experiences of discrimination),
lifestyle (smoking, alcohol use, recreational drugs, diet, exercise, perceived HIV
risk), childhood experiences (ACE), and socio-demographic questions (age, education,
marital status, race/ethnicity, household income, employment). Academic researchers
assisted with the development of hypotheses, analytic strategies, and statistical
analyses. In the subsections below, we describe only the specific instruments and
measures used to test our main hypotheses; however, the complete survey is available
as a supplemental file (Additional file 1).

Primary outcome: health-related quality of life

HRQoL was the primary study outcome and was measured using the CDC’s “Healthy Days
Measure” instrument, which is a validated scale used frequently in national health
surveillance surveys 28], 29]. Specifically, we used the brief version, referred to as Healthy Days Core Module
(4-items questionnaire), which captures the self-reported number of days in the past
30 days that individuals rated their physical or mental health as not good 30]. It includes the following components: 1) self-rated health, from poor to excellent
(ordinal); 2) number of days when physical health was not good during the past 30 days;
3) number of days when mental health was not good during the past 30 days; and 4)
number of activity limitations due to either physical or mental health illness (combined).
Total number of unhealthy days is then obtained by adding the responses to questions
2 and 3. Any sum greater than 30 was capped at 30, with a maximum of 30 unhealthy
days 30]. The outcome was operationalized through dichotomization of the total number of unhealthy
days, with poor HRQoL reflected by 14 or more total days. This cut-point has been
used by other authors as a discriminate measure to assess excessive unhealthiness
31]–33].

Primary exposure: adverse childhood experiences

To assess cumulative risk from childhood to adulthood, the main predictors included
participant recall of adverse events that occurred during the first 18 years of life.
We used the Brief Family History Questionnaire from the ACE study 6], a 10-item questionnaire collecting self-reported physical, emotional, and sexual
abuse, family dysfunction, and economic hardship. All items are dichotomous, yes/no
questions. An overall ACE score, which ranges from 1–10, is then calculated by adding
the number of “yes” responses. Higher scores have been found to be associated with
a wide range of adverse health outcomes, impaired quality of life, and higher mortality
7], 11], 34].

Potential mediators: smoking, alcohol use, diet, physical activity, and sleep disturbances

We hypothesized that contemporary lifestyle risk and protective factors of participants
could mediate the association between ACE and adult HRQoL. Lifestyle and behavioral
questions were obtained from multiple validated instruments that are frequently used
in health services research, which were adapted using input from community members.
Smoking was assessed with a question from the 2013 Behavioral Risk Factor Surveillance System (BRFSS)
35]: “Have you smoked at least 100 cigarettes in your entire life?” This question was
chosen by CAB members as being easier to respond to than a detailed frequency question
36]. The frequency of alcohol use per month was measured with a numerical question, also adapted from the 2013 BRFSS:
“In a typical month, how often do you drink alcoholic beverages?”

Although self-reported measures of dietary patterns are vulnerable to measurement
bias 37], 38], the CAB members felt that asking a few questions about diet was still an important
aspect to include in the survey as a general, non-sensitive behavioral item. Thus,
the survey included a short series of questions about intake of fruits and vegetables,
fatty, sugary, salty foods, and caffeinated drinks. Participants were asked to indicate
how often they consumed fruits and vegetables in a typical month, and we grouped responses
as: “once a month or less” coded as 1, “2–3 times a month” coded as 2.5, “once a week”
coded as 4, “2–3 times a week” equivalent to 10, and “4–6 times a week or more” coded
as 20. For this study, we focused on the consumption of fruits and vegetables as key protective factor because of stronger evidence regarding fruits and vegetable
self-reported measures. Specifically, brief instruments for fruit and vegetable intake
assessments have been found to be adequate for estimating relative risks in the relationship
between fruit and vegetable intake versus disease 39]. Because the survey included questions not previously validated, we consider our
approach a conservative one. Moreover, because we did not find that any of our self-reported
dietary measures were significantly associated with HRQoL in bivariate preliminary
analyses, we did not explore dietary items in greater detail in mediational analyses.
Sleep disturbances were also measured with a question we derived from the BRFSS 28]: “During the past 30 days, for about how many days have you felt you did not get
enough rest or sleep?” Physical activity was measured with one question that assessed physical activity, also from the BRFSS
28]: “During the past month, about how many days per week did you exercise for recreation
or to keep in shape (activities that make you sweat)?”

Stress appraisal

Cognitive appraisal of stress was measured with the 4-item Perceived Stress Scale,
which is a validated instrument used to make comparisons of subjects’ perceived stress
related to current events 40]. Questions are 5-point Likert type scaled (i.e., strongly disagree?=?1, to strongly
agree?=?5). A composite score was derived by adding the original scores and multiplying
by a factor of 5, which results in a 100-point scale. The higher the score, the higher
the risk for clinical psychiatric disorder 41].

Social support

Perceived social support was measured with five questions from the Medical Outcomes
Study Social Support Survey 42]. Using 5-point Likert type scales (‘None of the time’?=?1 to ‘All of the time’?=?5),
individuals were asked to indicate how often the following types of supports were
available to them if needed: 1) someone to confide in or talk about yourself or your
problems; 2) someone to share your most private worries and fears with; 3) someone
to help you if you were confined to bed; 4) someone to prepare your meals if you were
unable to do it yourself; and 5) someone to get together with for relaxation. Question
scores were added to yield a total social support score. To facilitate interpretation
in the community setting, the original scores ranging from 4 to 20 were converted
to a 100-point scale by multiplying the original score by five.

Sociodemographic confounders

Socio-demographic characteristics were collected from participants and included: age
in years (categorized as 18–35; 36–45; 46–55; and???56 years), sex, education (high
school vs. less than high school graduate or equivalent), current marital status,
race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, other), employment
status (employed, unemployed but able, and unable to work), annual household income
(categories: US$ 0–20,000; 20,001–40,000; and???40,000), and residential time in the
five target zip codes (5 years or less vs. more than 5 years).

Neighborhood confounders

Community resources, community-wide issues, and neighborhood cohesion levels were
considered as potential confounders in our analyses. First, resources available in
the community were measured with the following question: “Which of the following resources
are available to you in your community?” A list of resources was compiled using CAB
input and provided with the survey. Each item listed was weighted equally and the
final variable used in analyses was the total number of assets or resources reported.
Similarly, perceived community-wide issues were measured with one question on neighborhood
problems 43]: “Which of the following is a problem in the neighborhood?” Again, a list was developed
by the CAB, with the allowance for write-in entries. The final variable for analysis
consisted of the total number of different issues reported. Lastly, neighborhood social
cohesion was assessed by measuring the respondent’s level of agreement (on a 5-point
Likert type scale) with a set of questions proposed by Cagney and colleagues 44]: 1) “People around here are willing to help their neighbors”; 2) “This is a close-knit
neighborhood”; 3) “People in this neighborhood can be trusted”; and 4) “People in
this neighborhood generally don’t get along with each other” (reverse coded). These
questions were then summed to provide a total score, where higher scores indicated
higher neighborhood social cohesion.

Technology

The droidSURVEY software was used to design and administer the survey 45], which was installed on ten Hewlett-Packard Slate 7” portable tablet computers running
the Androidâ„¢ 4.2.2 operating system 46]. For Spanish-speaking participants, all questions were translated to Spanish by native
speakers and assessed for accuracy of translation using back translation and pilot
testing. The use of tablet-administered surveys allowed for a portable, convenient
means of data collection, and pilot testing revealed that community members found
touchscreen technology to be intuitive, easy to use, and enjoyable. Supervised by
the principal investigator, the study coordinator trained community members in the
use of tablets and survey implementation, the informed consent process, management
of tablet computers in the field, and how to assist participants with questions about
the study. Training consisted of three 2-hour sessions in the preceding month to the
survey.

Statistical analysis

The study population was described using descriptive statistics that included frequencies
and percentages for categorical variables and means and standard deviations for numerical
variables. Two-sided tests of equality for column proportions (z-tests for column proportions or t-tests for column means) were conducted to assess differences by outcome group. Tests
assumed equal variances and were adjusted for multiple comparisons using a Bonferroni
correction. All analyses were conducted with the IBM SPSS Statistics for Windows,
Version 22.0 (IBM Corp, Armonk, NY). Statistical significance was assessed at the
0.05 level.

Under the LCP framework, we posited that numerous accumulating risks and protective
factors mediate the association between ACE and the HRQoL proxy. As a first step,
separate logistic regression models were run to assess the independent effects of
life course social determinants on the outcome (?14 unhealthy days), adjusting for
individual socio-demographic and neighborhood covariates. We explored the role of
the following factors: stress, sleep disturbances, smoking, alcohol use, physical
activity, fruits and vegetable intake, and social support. The purpose was to identify
significant independent associations with HRQoL that could represent potential mediating
pathways in the relationship between ACE and HRQoL.

After a set of factors was identified as possible mediators, the next step was to
test for mediation or indirect effects 47]. The following three conditions must be established to determine whether mediation
had occurred (i.e., indirect effects) 48]: 1) that the independent variable (ACE) predicts the dependent variable (HRQoL),
2) that the independent variable (ACE) predicts the mediators (Ms), 3) that the mediator
(M) predicts the dependent variable (i.e., ?14 unhealthy days). Additionally, the
effects of confounders were considered 49].

Accordingly, mediation analyses for the dichotomous outcome (Y: HRQoL) were conducted
with logistic regression models to estimate the path coefficients in a two-mediator
model (X: ACE, M1: stress, M2: sleep disturbance). We controlled for socio-demographic
covariates (i.e., age, gender, race/ethnicity, no high school, and household income)
and neighborhood level factors (i.e., neighborhood cohesion, community resources,
and neighborhood issues). The first step was the estimation of controlled direct effects
through a series of logistic regressions. This was followed by the decomposition of
effects into total, indirect, and direct effects 50]. The formula used to calculate indirect effects for both mediators was: c?=?c’?+?(ab); where c is total effect, c’ is the direct effect, and a*b is the indirect effect. The indirect effect (ab) is the measure of the amount of
mediation, and equals the reduction of the effect of the causal variable on the outcome
or: ab?=?c – c’. For this purpose, we used an SPSS macro with asymptotic and resampling
strategies for comparing indirect effects in multiple mediator models, which included
bootstrapping to estimate confidence intervals for total and specific indirect effects
47], 51]. Missing values were handled with default option in SPSS for logistic regression
procedure, which is listwise deletion.