Temperament, Childhood Illness Burden, and Illness Behavior in Early Adulthood

Abstract

Objective:

Illness behaviors—or responses to bodily symptoms—predict individuals’ recovery and functioning; however, there has been little research on the early-life personality antecedents of illness behavior. This study’s primary aims were to evaluate: 1) childhood temperament traits (i.e., emotionality and sociability) as predictors of adult illness behaviors, independent of objective health; and 2) adult temperament traits for mediation of childhood temperament’s associations.

Methods:

Participants included 714 (53% male; 350 adoptive family and 364 control family) children and siblings from the Colorado Adoption Project (CAP; Plomin & DeFries, 1983). Structural regression analyses evaluated paths from childhood temperament to illness behavior (i.e., somatic complaints, sick days, and medication use) at two adulthood assessments (CAP years 21 and 30). Analyses controlled for participant age, sex, family type (adoptive or control), adopted status, parent education/occupation, and middle childhood illnesses, doctor visits, and life events stress.

Results:

Latent illness behavior factors were established across two adulthood assessments. Multilevel path analyses revealed that higher emotionality (fearfulness) in adulthood—but not childhood temperament—predicted higher levels of illness behavior at both assessments. Lastly, lower emotionality-fearfulness partially mediated the effect of higher childhood sociability on adult illness behavior.

Conclusions:

Results suggest the importance of childhood illness experiences and adult emotionality (fearfulness) in shaping illness behavior in early adulthood. They also suggest a small, protective role of childhood sociability on reduced trait fearfulness in adulthood. These findings broaden our understanding of the prospective links between temperament and illness behavior development, suggesting distinct associations from early-life illness experiences.

In 2013, United States healthcare expenditures reached $2.9 trillion, with an average personal health cost of $9,255 per capita (National Center for Health Statistics, 2014). Such daunting expenditures point to a need for increased efficiency in the delivery and utilization of health services. As a first step, however, the process of illness must be better understood. In others words, what psychological and behavioral processes occur before people seek (or choose not to seek) formal health services? Illness behavior—a psychosocial construct defined as individuals’ perceptions, evaluations, and responses to symptoms that signify illness (Mechanic, 1962)—provides a framework for examining who is more likely to react to bodily sensations, and under what circumstances.

Illness behaviors are associated with individual health outcomes, such as reported pain levels, disease functioning (Schüssler, 1992; Crane & Martin, 2002), timely detection of life-threatening illnesses (e.g., cancer, Van Osch et al., 2007), and return to work (Broadbent, Ellis, Gamble, & Petrie, 2006; Keefe, Crisson, Maltbie, Bradley, & Gil, 1986). Examples of illness behavior range from denial of symptoms to symptom monitoring and preoccupation; from stoicism to expression of negative affect (e.g., complaining, grimacing); and from delayed medical care-seeking to care-seeking for minor physical complaints, or absenteeism from work and avoiding social obligations. Because illness behavior describes not only how people respond to their symptoms, but also what they choose not to do (i.e., delaying treatment), both extremes of responding must be targeted for optimizing health care delivery. Illness behaviors are also distinct from primary health behaviors: Whereas primary health behaviors refer to the preventative actions of healthy or symptomless individuals (e.g., diet, screenings), illness behaviors encompass individuals’ observable responses—or lack there-of—to bodily symptoms and self-appraisals as “ill”. Illness behavior is not only a function of biological risks, but it is also influenced by psychosocial factors. Access to healthcare, cultural norms, prior experiences, social support, and personality, along with symptom qualities (e.g., ambiguity, visibility), account for variability in illness behaviors (e.g., Egan &Beaton, 1987; Hagger & Orbell, 2003; Mechanic, 1995; Shannon, 1977). Most studies on illness behavior, however, have examined one aspect at a time (e.g., symptom reports, health utilization) and have considered cross-sectional or short-term associations with psychosocial factors in adulthood (Sirri, Fava, &Sonino, 2013). Furthermore, although prior longitudinal work has examined associations of childhood personality and illness with adults’ self-rated (Hampson, Goldberg, Vogt, &Dubanoski, 2007) and objective health (Kubansky, Martin & Buka, 2009), these early-life factors’ associations with adult illness behaviors as a unifying, multivariate construct has received little empirical attention. The present study examined middle-childhood illness burden and temperament as antecedents of a latent illness behavior factor in early adulthood, as well as mediation by adult temperament traits.

Personality development and illness behavior

The four most common amplifiers of bodily symptoms—attention, mood, beliefs, and social circumstances (e.g., interpersonal conflict, stress)—are entirely psychosocial in nature (Barsky, 1988). Because the construct of personality theoretically encompasses stable individual differences in the attention, mood, and beliefs that shape behavior, it represents a promising target for research on illness behavior development. In particular, the Big Five trait of neuroticism is posited to influence the ways that symptoms are perceived and labeled (Costa & McCrae, 1987; Leventhal, Leventhal, & Contrada, 1998), and these perceptions, in turn, predict illness responses. For example, greater neuroticism is associated with an increased internal bodily focus (Costa & McCrae, 1987), reduced internal locus of control in the face of health threats, and higher perceived vulnerability to disease (Gerend et al., 2004). From this research, it follows that people who score higher on neuroticism are more responsive to their symptoms. Conversely, the Big Five trait of extraversion might decrease illness behaviors, because highly extraverted individuals have more difficulty shifting their attention from external to internal stimuli (Pennebaker & Brittingham, 1982) and are less likely than introverts to accept illness-related restrictions on social activity. For example, individuals scoring high on extraversion are relatively more likely to complain to others about pain, yet they also report less distress and pain sensitivity (Harkins, Price, & Braith, 1989). Thus, extraverts might be more likely than introverts to engage their support systems when ill, but less likely to notice symptoms. Despite empirical support for concurrent associations of adult personality with a range of illness behaviors, the emergent relationships of these personality influences on illness behavior are not well understood (Crane & Martin, 2002; Schüssler, 1992). No studies to date have evaluated the predictive role of these traits in early-life on adulthood illness behavior development.

The Present Study

To understand the role of personality in the development of illness behavior, a lifespan approach is useful for examining when, and to what extent, individuals’ traits contribute over time. Thus, the current study leverages prospective data from the Colorado Adoption Project (CAP; Plomin & DeFries, 1983) to examine illness behavior as a developmental process, addressing the emergence of middle childhood temperament traits and illnesses as predictors of early adult illness behavior. The temperament traits of emotionality and sociability are viewed as moderately heritable, stable precursors of neuroticism and extraversion (Buss & Plomin, 2014; Goldsmith et al., 1987). Temperament more generally is viewed as a key aspect of personality and encompasses the afore-mentioned symptom amplifiers of attention and mood. Although personality traits are characterized by more specific beliefs and values, temperament reliably predicts these cognitions (Rothbart & Bates, 2006). Empirical findings on the links between temperament and Big Five personality traits in adulthood suggest two underlying common affective-motivational factors of extraversion and negative affect (Evans & Rothbart, 2007). Other longitudinal work predicting adult Big Five traits from middle childhood impulsivity and inhibition found that these two dimensions account for more than 30 percent of variability in adult personality (Deal, Halverson, Havill, & Martin, 2005). Thus, although this study focuses on temperament, it includes standard, reliable measures of two traits with demonstrated concurrent and prospective links with adult personality (Shiner & DeYoung, 2013). Furthermore, the exploration of temperament in middle childhood is of particular interest, as this is a time period in which children begin to differentiate themselves from others in terms of their psychological traits (Harter, 2012).

Apart from childhood personality, a widespread literature also shows that early childhood health has enduring associations with chronic disease and physical functioning in adolescence and early adulthood(Case, Fertig, & Paxson, 2005; Haas, 2008). These results are underscored by a theoretical framework of cumulative risks or ‘insults’ (Kuh & Ben-Shlomo, 2004), which posits that childhood biopsychosocial risk factors may accumulate across the lifespan to influence adult health and behavior; thus, the current study adjusted for parent reports of childhood illnesses to determine the extent to which child temperament associations remained after accounting for concurrently-measured, objective health. This study’s hypotheses were as follows: 1) Higher emotionality in childhood would predict higher levels of adult illness behavior; 2) Childhood sociability would also predict illness behavior, although there was no hypothesis of directionality given the mixed findings in the literature; and 3) Adult emotionality and sociability would partially mediate the relationship between childhood temperament and illness behavior in adulthood. Given the moderate stability of temperament across the lifespan (Buss & Plomin, 2014), the strength of childhood temperaments’ associations was expected to decrease once proximal adult temperament traits were added to the predictive model. Finally, greater childhood illness burden was expected to predict higher levels of adult illness behavior.

Method

Participants

Study participants were child members of adoptive families or matched control families from the Colorado Adoption Project (CAP) (Plomin & DeFries, 1983), begun in 1977. The CAP is a longitudinal, adoption study of genetic and environmental influences on behavioral development. The CAP provides rich, prospective data on behavioral development and health across a key developmental transition into early adulthood. Because clinical studies on illness behavior may be biased towards clinic-based samples who have already sought out healthcare (i.e., people who tend to fall along the higher extreme of the illness behavior continuum), the CAP therefore represents a unique opportunity to examine the lifespan development of this construct within a population-based sample of U.S. adults. CAP proband participants (adopted and matched control children) and their siblings were assessed almost annually from infancy to approximately 21 years of age (i.e., across CAP assessment years 1 to 21), with a more recent (year 30) assessment completed in 2011 including a subset of participants between 30 and 35 years of age. In the present study, the phrases “CAP assessment year” or “CAP year” refer to the measurement occasion, and do not necessarily reflect participants’ actual ages. Across CAP years 7–15, all participants completed assessments based on their current grade in school, beginning with first grade, such that all of the children were around the same age at a particular assessment year (i.e., in third grade at assessment year 9). At CAP year 16, participants completed the assessment as close to their 16th birthday as possible. Thus, there was not a wide range of ages at testing, and siblings were rarely administered the same tests during a single visit. Of particular relevance to this study, repeated measures of self-reported temperament were available across CAP assessment years 9–16 and 21, and illness behaviors were assessed at CAP assessment years 21 and 30.

The complete CAP sample consists of 493 families (247 adoptive; 246 control). Within adoptive families, there were both adopted and biological children. Adoptive and control families were matched on the adopted or control child’s gender, number of children, and the father’s age, education, and occupational status (see Plomin & DeFries, 1983). Adopted children were recruited from local social services in Colorado and placed into their adoptive homes, on average, 29 days after birth (range = 2 to 172 days). Prior to adoption, they received foster care (Rhea et al., 2013a). The majority of the CAP sample self-identifies as Caucasian (95% control parents, 90% adoptive parents) and the remaining as Hispanic/Latino or Asian-American. The sample was of slightly higher socioeconomic status compared to the U.S. average at the time CAP was initiated; however, its variability is comparable to U.S. norms (Rhea et al., 2012; 2013). Ethical approval for the CAP study was provided by the University of Colorado, Boulder and University of California, Riverside Institutional Review Boards.

Within these families, 714 adopted and matched control children and their siblings were included in the analysis sample from those invited to participate in the childhood CAP assessments (53% male; 350 from adoptive families and 364 controls).¹ Of the 350 participants from adoptive families, 30 were biological children and 320 were adopted. Participants were nested within 477 families, each including up to three siblings (51% single-child). Analyses included all individuals with demographic data (e.g., family type, age, sex), and substantive data (i.e., temperament, illness behavior) at the CAP year 9, year 21 or year 30 assessments. Of the 714 participants, 88% (n =625) had data at the year 9 assessment, 77% (n =551) had data at the year 21 assessment, and 39% (n= 275) had data at the year 30 assessment. Of the 625 individuals with data at year 9, 80% (n= 501) also had longitudinal data at year 21, and 40% (n =247) had longitudinal data at year 30. Some of the missing data at the adulthood assessments is the result of attrition; however, there were some assessments in which siblings of the original probands were not recruited to participate (Rhea et al., 2013b). For the year 30 assessment, a smaller subset of participants was randomly selected for recruitment due to funding constraints.

In terms of attrition analyses, those who participated at CAP year 21 had parents with, on average, higher education (t₍₇₀₈₎ = 2.57,p = .010) and occupational prestige(t₍₇₀₈₎ = 3.05,p= .002). Non-adopted participants and those from control families were also more likely to have data at year 21 (χ² (1) = 5.84, p = .016, and χ² (1) = 4.66, p = .031, N = 714, respectively). Data were assumed to be missing at random (MAR), and maximum likelihood estimation was applied which ensures the validity of results under this assumption. All predictive analyses were adjusted for parent reports of their child’s past-year illnesses, doctor visits, and life events stress, parents’ highest reported education and occupational status, as well as age, sex, family type, and adopted status. The emphasis for the current study was not to compare adoptees and non-adoptees, and we did not have reason to expect illness behaviors to substantially differ by adopted status. Nevertheless, we accounted for adopted status in statistical analyses to improve the accuracy of model parameters. We controlled for both family type and adopted status, because adoptive families also included biological (non-adopted) children.

Measures

Temperament (CAP years 9, 21).

Child temperament was measured using the Colorado Childhood Temperament Inventory (CCTI; Rowe & Plomin, 1977). The CCTI was derived from the EAS survey (Buss & Plomin, 2014), and measures four key temperament dimensions: Emotionality, Activity, Sociability, and Impulsivity (Plomin, Corley, Caspi, Fulker, & DeFries, 1998). The present study included the Emotionality and Sociability CCTI scales from CAP year 9. Repeated measures of child self-reported temperament, as measured with the CCTI, were included across CAP years 9–16. Year 9 was chosen because this was the first middle childhood wave in which children provided self-reports of their temperament.² Adult temperament was measured with the Emotionality-Activity-Sociability Temperament/-Impulsivity Survey (EAS-I; Buss & Plomin, 1975). The EAS-I further divides emotionality into two subscales: Fearfulness and Anger. The present study included Emotionality-Anger, Emotionality-Fearfulness, and Sociability EASI scales from CAP year 21. Scales were previously validated and show good reliabilities (CCTI subscales, median α = .80; EASI, median α = .72; see Hubert, Wachs, Peters-Martin, & Gandour, 1982).

Illness behavior (CAP years 21, 30).

All illness behavior measures were gathered at CAP years 21 and 30 as part of a larger telephone-based questionnaire. Illness behavior was operationalized as any measure reflecting participants’ evaluations and responses to bodily symptoms, regardless of whether these physical complaints were corroborated by an objective health measure. This contrasts with physical health, which refers to participants’ physical wellbeing, or the absence of bodily symptoms and any acute or chronic health conditions. Measures included in this study are discussed in detail below.

Somatic complaints.

The first indicator of illness behavior was a checklist of bodily symptoms, reflecting participants’ disease or symptom preoccupation (c.f., Illness Behavior Questionnaire/IBQ; Pilowsky & Spence, 1975). In the current study, participants indicated how often they experienced 9 common symptoms with no definable diagnoses (e.g., dizziness, nausea, stomachaches), each on a 6-point scale from never (1) to daily (6). A confirmatory factor analysis constrained loadings to be equal across waves without loss of fit; this yielded an invariant “somatic complaint” factor (see Supplement).

Sick days.

This item indexed participants’ investment in the sick role, asking their “frequency of missed school/work due to illness” on a 6-point scale from never (1) to daily (6).

Medication use.

This self-report item assessed participants’ “frequency of taking medication for emotion/nerve problems”, on a 6-point scale from never (1) to daily (6). This item indexed participants’ somatic (rather than psychological) orientation toward illness, denial of life stresses, and affective disturbance (general anxiety) (c.f., Illness Behavior Questionnaire/IBQ; Pilowsky & Spence, 1975).

Control Variables

Parent-report data on participants’ past-year illness burden, number of doctor visits, and life event stress from the CAP year 9 assessment, the highest-reported parental education and occupational prestige, as well as participants’ actual age at each CAP assessment year (9, 21,30), sex (−.50=male; +.50=female), family type (−.50=adoptive; +.50=control), and adopted status (0 =non-adoptee;1 =adoptee), were examined as potential covariates. Illness burden was a composite of parents’ reports of the total frequency of any health problems their children had experienced in the past year: i.e., sum of International Classification of Diseases and Related Health Problems symptom codes (ICD-9;http://www.icd9data.com/2011/Volume1/default.htm), each multiplied by frequency of occurrence. Previous research suggests that parent report of children’s doctor visits and medical chart reviews are generally in strong agreement (Craig, Cox, & Klein, 2002). Past-year life events stress was measured via the Social Readjustment Rating Scale (SSRS; Coddington, 1972). Parents indicated which of 33 life events had occurred during the previous year and how upsetting the event was for their child from 0 (not at all) to 3 (very much). These ratings were then summed. Participants’ exact age at each CAP assessment year were entered to account for possible age effects and the distance between measurement occasions. These variables were centered at ages 9, 21, and 30 years, respectively. Two variables were created reflecting adoptive or control parents’ highest reported levels of education and occupational attainment (National Opinion Research Center/NORC scores) at CAP intake and year 7 assessments (i.e., the highest scores between the two assessments, regardless of whether they came from the mother or father).

Statistical Procedures

Model-fitting was conducted using Mplus (Version 8, Muthén & Muthén, 2012). A confirmatory factor analysis (CFA) using robust weighted least squares estimation (WLSMV) was conducted to test and validate a factor indexed by somatic complaints, staying home from school or work due to illness, and medication use (see Supplement). All models accounted for data dependency (i.e., nesting within families). The final illness behavior factor included somatic complaints (factor score), sick days, and medication use.

The primary analyses applied multilevel, latent path regression models using robust maximum likelihood estimation (MLR). Full-information modeling of all data was applied to reduce any possible attrition bias. Models were evaluated at both adulthood assessments (CAP years 21 & 30). In the first step(Model 1), illness behavior was regressed on year 9 emotionality and sociability, adjusting for covariates, and the fit was evaluated with a comparison model in which the two paths between year 9 emotionality and sociability to adult illness behavior were dropped, but otherwise identical to Model 1. Next, we fitted a mediation model(Model 2)with year 21 emotionality and sociability traits added to Model 1 as mediators of child temperament traits on adult illness behavior (see Figure 1), and compared its fit with three nested models, in which: a) the two paths from year 9 temperament traits to adult illness behavior were dropped; b) the six paths from child to adult temperament traits were dropped; and c) the three paths from adult temperament traits to illness behavior were dropped, but all otherwise identical to Model 2. Nested models were compared using the likelihood ratio test formula specified by Muthén and Muthén (2010) (http://www.statmodel.com/chidiff.shtml) for MLR estimation, and mediation was evaluated with the MODEL CONSTRAINT command. Grand-mean centering was used for all predictors, except stressful life events and demographics. Age was treated as a covariate, where year 9 age was regressed out of all outcomes (i.e., adult temperament, illness behavior), and age at year 21 or year 30 was regressed out of illness behavior (see Supplemental figure).

Figure 1.

Path regression models of Year 21 illness behavior: A) Model 1, illness behavior regressed on Year 9 emotionality, sociability, and illness burden; B) Model 2, mediation model of illness behavior via Year 21 adult emotionality (fearfulness, anger) and sociability. Dashed lines = tested mediation paths. Covariates (not shown): parent education and occupational attainment, child sex, adopted status, family type; Year 9 age, doctor visits, and life events stress and Year 21 age. Somatic= somatic complaints; Sick_days= frequency of staying home from school or work due to illness; Med_use= frequency of medication use for emotion/nervous problems. * p < .05. ** p < .01. *** p < .0001.

Results

Descriptive analyses are presented in Table 1. The average age of participants across the three assessments was 9.48 years at CAP year 9 (SD = 0.37), 21.54 years at CAP year 21(SD =0.74), and 31.86 years at CAP year 30(SD =1.28), respectively. Natural logarithmic transformations were applied to childhood doctor visits and illness burden variables to adjust for significant positive skew, and reported pairwise correlations (Table 2) were based on transformed values. Correlations between child temperament and observed adult illness behavior items were modest to small (rs ranged from −.12 to .12). Notably, child emotionality was positively associated with year 21 somatic complaints (r(468) = .12, p = .008), whereas child sociability was negatively associated (r(466) = −.12, p = .010). Stronger associations emerged for the year 21 temperament dimensions (rs ranged from-.15 to.40). Higher emotionality-fearfulness at year 21 was positively associated with all three illness behavior indicators at CAP year 21(n = 518 to 533; rs = .18 to .40, all ps < .001), and with somatic complaints and medication use at CAP year 30 (r (258) = .19, p = .002, and r (260) = .24, p < .001, respectively). Higher emotionality-anger at year 21 was positively associated with year 21 somatic complaints (r (518) = .12, p =.009), whereas higher sociability at year 21 was negatively associated with both year 21 somatic complaints and frequency of medication use (r (518) = −.13, p = .003; and r (533) = −.15, p = .001, respectively). Correlations between child and adult temperament traits were small. Child emotionality was positively associated with adult emotionality-fearfulness (r (489) = .13, p = .005) at CAP year 21. Child sociability positively correlated with adult sociability (r (487) = .17, p < .001), and negatively correlated with adult emotionality-fearfulness (r (487) = −.18, p < .001) at CAP year 21.

Table 1.

Summary of Descriptive Statistics for Temperament, Illness Behavior, and Covariates

Quantitative variable	N	Mean (SD)	Min	Max	Categorical variable	Univariate proportions (%) and counts (n)
Temperament						Response category
Y 9 emotionality	622	10.60 (4.03)	4.00	20.00		Never	< Once a year	Once a year	Once a month	Once a week	Daily
Y 9 sociability	621	14.83 (3.36)	4.00	20.00	Y 21 sick days	25.0 (133)	27.8 (148)	38.9 (207)	7.9 (42)	0.2 (1)	0.2 (1)
Y 21 emotionality- fear	539	11.48 (3.39)	5.00	22.00	Y 21 medication use	87.5 (467)	3.4 (18)	1.7 (9)	1.1 (6)	0.2 (1)	6.2 (33)
Y 21 emotionality- anger	539	12.30 (3.49)	5.00	25.00	Y 30 sick days	28.4 (77)	30.6 (83)	35.4 (96)	5.2 (14)	0.4 (1)	0 (0)
Y 21 sociability	539	18.51 (4.02)	5.00	25.00	Y 30 medication use	79.3 (215)	5.5 (15)	1.1 (3)	1.1 (3)	0.7 (2)	12.2 (33)
Covariates
Y 9 illness history (raw)	545	3.41 (2.75)	1.00	20.00
Y 9 illness historya	545	0.98 (0.69)	0.00	3.00
Y 9 doctor visits (raw)	545	2.47 (2.88)	0.00	20.00
Y 9 doctor visitsa	545	0.98 (0.71)	0.00	3.05
Y 9 stress	557	4.83 (3.55)	0.00	23.00
Y 9 age	625	9.48 (0.37)	8.58	10.92
Y 21 age	551	21.54 (0.74)	20.83	25.75
Y 30 age	274	31.86 (1.28)	30.00	35.44
Parent education	710	16.43 (2.17)	12.00	22.00
Parent occupation	710	58.02 (10.43)	22.00	81.00
Illness behavior
Y 21 somatic (factor)	518	0.01 (0.85)	−1.48	3.99
Y 30 somatic (factor)	267	0.02 (0.86)	−1.45	2.73

Note.

^aLog-transformed variable.

Table 2.

Zero-Order Correlations: Observed Illness Behavior Items, Temperament, and Covariates

	Y 21 somatica	Y 21 sick daysb	Y 21 medb	Y 30 somatica	Y 30 sick daysb	Y 30 medb
Temperament
Y 9 emot	.12**	−.02	.04	−.01	.08	.07
Y 9 soc	−.12*	−.02	−.06	− .04	−.11	−.11+
Y 21 emot-F	.40***	.18***	.26***	.19**	.10	.24***
Y 21 emot-A	.12**	.06	.07	.10	.01	.01
Y21soc	−.13**	−.07+	−.15***	−.03	−.02	.04
Covariates
Sexc (.50=Female)	.28***	.09*	.13**	.25**	.18**	.18**
Family typec (.50=Control)	−.05	.01	−.03	−.02	−.04	−.05
Adoptedc (1 = Adoptee)	.01	−.01	.02	.01	.03	.09
Y 9 illnessesd	.13**	.07	.04	.22**	.11	.09
Y 9 doctor visitsd	.02	.05	−.04	.14*	.01	−.05
Y 9 stress	.01	.03	.02	.03	−.09	−.04
Parent educ	−.03	.01	.01	−.02	.12*	.03
Parent occup	−.01	.02	.03	.01	.08	.06
Y 9 age	−.01	−.04	.00	−.05	−.08	.03
Y 21 age	−.10*	−.10*	−.07	.08	−.06	−.07
Y 30 age	.07	.07	−.10	.03	.03	.07

Note. All were pairwise zero-order correlations. Ns ranged from 405 to 533 (Year 21) and from 210 to 270 (Year 30). All p values were two-tailed. emot-F = emotionality-fearfulness; emot-A = emotionality-anger.

^aPearson-product correlation coefficients.

^bSpearman correlation coefficients for rank-ordered categorical variables.

^cEffects-coded variable.

^dLog-transformed variable.

⁺p < .10.

^*p < .05.

^**p < .01.

^***p < .001.

Multilevel Latent Path Regressions

For CAP assessment year 21, Model 1 evaluated child temperament traits’ prediction of adult illness behavior (see Figure 1, Table 3, Model 1 results; see Supplement for full results). As a set, child temperament traits were significant at the .05 level (χ² (2) = 7.369, p = .025), as these paths could not be dropped without reducing model fit. Higher child emotionality did not uniquely predict higher levels of illness behavior (β = .06, p = .195), nor did child sociability (β = - .07, p = .149). Notably, greater childhood illness burden predicted a higher loading on adult illness behavior (β = .15, p = .005). Despite some debate in the methodological literature on whether mediation analyses should progress in the case of non-significant paths from predictor-to-outcome (e.g., the causal-steps approach; Baron & Kenny, 1986), more recent work has proposed that significant mediation can still occur with a non-significant overall effect (MacKinnon & Fairchild, 2009). For example, factors such as reduced power to detect a small effect that has been reported with the causal-step approach, or errors in the statistical test itself can contribute to small or non-significant overall relationships (MacKinnon & Fairchild, 2009). Therefore, the present study proceeded with mediation models, adding the prediction of adult temperament traits on illness behavior. In Model 2, year 21 emotionality (fearfulness and anger subscales) and sociability were added to test for mediation between child temperament and illness behavior (see Figure 1, Table 3, Model 2 results). To evaluate the fit of Model 2, the two paths from childhood emotionality and sociability to illness behavior were dropped in a nested model, which did not significantly reduce the model fit (χ² (2) = 1.145, p = .564). However, dropping the paths from child temperament to adult temperament traits (i.e., the six paths from child emotionality and sociability to adult emotionality-fearfulness, emotionality-anger, and sociability) in another nested model resulted in significantly reduced model fit (χ² (6) = 27.029, p < .001). Thus, although child emotionality and sociability did not predict illness behavior, its associations with adult temperament remained substantial. Lastly, dropping the three paths from adult temperament traits to year 21 illness behavior significantly reduced the model fit (χ² (3) = 87.992, p < .001), suggesting that, as a set, these traits were significant predictors.

Table 3.

Adult Illness Behavior Model Results (Years 21 and 30): Standardized Estimates

		Year 21				Year 30
		Model 1		Model 2		Model 1		Model 2
Outcome		B	SE	B	SE	B	SE	B	SE
Illness Behavior	Emot 9	0.06	0.05	0.04	0.05	−0.02	0.09	−0.03	0.08
	Soc 9	−0.07	0.05	−0.02	0.05	−0.10	0.09	−0.10	0.09
	Illness 9	0.15**	0.05	0.12*	0.06	0.27**	0.10	0.28**	0.10
	Doc 9	−0.05	0.06	−0.03	0.06	−0.03	0.11	−0.03	0.12
	Stress 9	−0.03	0.05	−0.03	0.05	−0.06	0.09	−0.07	0.09
	Parent education	−0.03	0.06	−0.02	0.06	0.02	0.11	0.04	0.10
	Parent occupation	0.02	0.06	0.00	0.06	0.06	0.08	0.04	0.08
	Age 9	0.07	0.05	0.07	0.05	0.01	0.09	0.03	0.10
	Sex	0.28***	0.04	0.16**	0.05	0.34***	0.08	0.29**	0.09
	Family type	−0.14	0.11	−0.17	0.12	−0.01	0.22	0.04	0.23
	Adopted status	−0.10	0.11	0.04	0.10	0.03	0.23	0.02	0.24
	Emot-F 21	--	--	0.36***	0.06	--	--	0.16	0.11
	Emot-A 21	--	--	0.02	0.05	--	--	0.07	0.07
	Soc 21	--	--	−0.07	0.05	--	--	−0.01	0.08
	Age 21	--	--	−0.09*	0.04	--	--	0.07	0.09
	Age 30	--	--	--	--	--	--	0.11	0.09
Year 21 Temperament	Emot 9 → Emot-F	--	--	0.02	0.04	--	--	0.02	0.04
	Emot 9 → Emot-A	--	--	0.09+	0.05	--	--	0.09+	0.05
	Emot 9 → Soc	--	--	−0.04	0.05	--	--	−0.04	0.05
	Soc 9 → Emot-F	--	--	−0.13**	0.04	--	--	−0.13*	0.04
	Soc 9 → Emot-A	--	--	0.09+	0.05	--	--	0.09+	0.05
	Soc 9 → Soc	--	--	0.15**	0.05	--	--	0.15**	0.05
Model fit indices
AIC/BIC		25419.3/25890.3		35211.2/ 35965.6		24933.4/ 25463.8		34781.2/ 35599.6

Note. Doc = doctor visits; Emot = Emotionality (-F = Fear, -A = Anger); Soc = Sociability.

⁺p < .10.

^*p < .05.

^**p < .01.

^***p < .001.

Bold values = statistically significant estimates.

In Model 2, the effect sizes for child temperament dropped in magnitude and significance, but maintained their directions. Year 21 emotionality-fearfulness was a strong predictor of year 21 illness behavior (β = .36, p < .001). Childhood illness burden remained a significant predictor of year 21 illness behavior (β = .12, p = .023), independent of associations with concurrent adult temperament traits. Modest associations between temperament traits in childhood and adulthood (CAP year 21) were also observed, such that child sociability positively predicted adult sociability (β = .15, p = .001) and negatively predicted adult emotionality-fearfulness (β = - .13, p =.003). Child emotionality’s associations with adult emotionality-anger (β = .09, p = .068) and emotionality-fearfulness (β = .02, p = .594) were non-significant at the .05 level. Mediation analyses suggested a small, significant indirect effect of year 21 emotionality-fearfulness (b = - .01, 95% CI [−.016, −.004], p = .010), such that reduced fearfulness partially mediated the association of greater child sociability with lower adult illness behavior scores. The direct and total effects were non-significant (ps > .10).

For CAP assessment year 30, models again evaluated the childhood antecedents of an adult illness behavior factor and mediation by adult temperament traits (see Supplement for full results). Results revealed that, again, child emotionality and sociability did not predict illness behavior at assessment year 30, although childhood illness burden remained a significant predictor (β = .28, p = .005). Year 21 emotionality-fearfulness did not significantly predict higher illness behavior scores (β = .16, p = .135). Associations between child and adult temperament traits remained unchanged from the year 21 models, with greater sociability in childhood predicting both higher sociability and lower emotionality-fearfulness in adulthood.

Discussion

The present study evaluated the extent to which middle childhood temperament traits predict adult illness behaviors assessed over a decade in the CAP. In this young adult sample, somatic complaints, sick days, and medication use mapped well onto the theoretical construct of illness behavior originally proposed by Mechanic (1962; 1978) and adapted by others (e.g., Egan & Beaton, 1987; Prior & Bond, 2008; Sirri et al., 2013; Wyke, Adamson, Dixon, & Hunt, 2013). Although small zero-order correlations were evident, child temperament did not significantly predict illness behavior in the multilevel path analyses. Adult temperament (emotionality-fearfulness), however, did predict higher levels of concurrent illness behavior, suggesting the importance of proximal, emotional trait influences. Furthermore, childhood illness burden directly predicted higher illness behavior at both adult assessments, independent of adult temperament traits and all other childhood covariates. The hypothesis that child emotionality and sociability would directly predict higher illness behavior in early adulthood, after accounting for these same traits in adulthood, was not supported. Adult emotionality-fearfulness, however, partially mediated the relation between child sociability and illness behavior. Behavioral inhibition in childhood is posited to be one indicator of fearful temperament, and persistently high levels throughout adolescence predict negative adjustment outcomes (e.g., social anxiety; Chronis-Tuscano et al., 2009). This study’s finding that the fear component of trait emotionality (as opposed to anger, a more externalizing regulatory process) predicted greater responsiveness to symptoms is supported by social developmental research on the differential regulatory processes of anger and fear (Cole, Martin, & Dennis, 2004). Furthermore, research on healthcare over-utilization suggests emotional concerns (e.g., psychological distress, depression) predict frequent attendance in primary care (Kessler & Stafford, 2008). Thus, our finding might underscore trait-based emotional processes as amplifiers of symptoms and people’s preferences for responding. Alternatively, because adulthood temperament traits and illness behavior were assessed at the same measurement occasion (CAP year 21), another possible interpretation of these findings is that greater responsiveness to physical symptoms predicted increased fearfulness.

Whereas adult emotionality-fearfulness predicted more responsiveness to symptoms as described, adult sociability did not significantly predict illness behavior. Higher sociability in childhood, however, predicted lower adult emotionality-fearfulness scores, suggesting a small, indirect path from child sociability to early adult illness behavior. The small effect size is generally consistent with longitudinal or concurrent studies of temperament-personality associations (e.g., Evans & Rothbart, 2007; Martin & Friedman, 2000; Hagekull & Bohlin, 2003). Of greatest note, higher childhood illness burden predicted higher levels of adult illness behavior. The non-significant relationship between childhood illness burden and adult temperament suggests that the effect of childhood health events on adult illness behavior might be direct, rather than mediated by personality processes. Although extant research on associations of childhood health with adult outcomes has primarily focused on objective health indices, other work suggests the cognitive and emotional aspects of illness—captured in illness behavior—might also have long-term associations with health. For example, Walker and colleagues (2012) found that children with an abdominal pain response profile including low coping efficacy, high catastrophizing, negative affect, and activity restrictions were more likely than other subgroups to subsequently meet clinical criteria for a gastrointestinal disorder in adolescence and early adulthood (Walker, Sherman, Breuhl, Garber, & Smith, 2012). Therefore, qualitative aspects of these childhood health conditions might warrant further exploration in relation to stability or changes in adult health and illness behavior.

Limitations and Future Directions

Strengths of the present study include its large, representative sample, prospective study design, and robust statistical approach. This study’s inclusion of a relatively healthy sample of adults was another strength, because examining common symptoms that everyone experiences allowed for greater individual variability in the illness behavior outcome of interest. Most participant samples in this area of research have been recruited from clinics or hospitals, biasing estimates toward those individuals who have already sought out professional treatment, while ignoring the large portion of the population that is not as likely to seek care. Although the present study demonstrated that illnesses as early as middle childhood and emotionality-fearfulness in early adulthood predicted adult illness behavior across a decade in the CAP, the role of the social context in accounting for individual variability cannot be ignored. For example, living in a rural as compared to an urban environment with limited access to services has been shown to delay the initial recognition and care seeking for any recognized set of symptoms (Shannon, 1977). These structural features combined with sociocultural differences in norms regarding help-seeking or expressing pain (Shannon, 1977), might partially underlie health disparities in a variety of domains. The CAP sample was largely Caucasian, and the range of SES, while representative of the community from which it was recruited, did not include the lowest levels of SES. The results of our attrition analysis suggested that non-adopted participants and those with parents of higher educational and occupational attainment were more likely to provide outcome data at the CAP year 21 assessment. These factors did not, however, predict participation at CAP assessment year 30, and the consistency of our findings across adulthood assessments lends support to the internal validity of these results. Although we used robust statistical methods, such as adjusting for these observed confounders and applying maximum likelihood estimation, to the extent that the present study did not capture the full range of SES or did not account for unobserved confounders, this may have biased our results. Prior work also suggests the intergenerational transmission of illness behaviors like symptom reporting and medication use (Levy, 2010). Future research examining the mechanisms of parent-child illness behavior associations is therefore warranted. Additional limitations of the present study are that the illness items included were not part of a previously validated scale; however, invariance in a latent illness behavior factor was established. Causal interpretations are limited, but the use of prospective, longitudinal data lends plausibility to downstream effects of childhood illness on adult illness behaviors.

Conclusions

The present study suggests that dispositional negative affect (i.e., fearfulness) is proximally associated with adult illness behaviors—specifically, those related to increased illness responding (i.e., more frequently reporting symptoms, staying home from school and work, and use of medications). Importantly, the study’s recruitment of a normative (non-clinical) young adult sample lends support to the notion that everyday experiences of fear are associated with higher levels of responding to one’s symptoms. Thus, emotional processes might be an important target in in efforts to improve patient health and healthcare delivery. Although some preliminary evidence suggests that early traits of emotional instability serve as risk factors for psychiatric disorders like depression and anxiety, with substantial genetic overlap (Hettema, Neale, Myers, Prescott, & Kendler, 2006), the present study cannot make claims about the role of mental health in illness behavior. However, the role of emotional processes more broadly as a determinant of illness behavior has been acknowledged in previous illness behavior models (e.g., the Common Sense Model; Leventhal et al., 1998), but remains poorly defined (Martin & DiMatteo, 2013; Wyke et al., 2013). Thus, future work would benefit from defining and directly assessing trait-based emotional processes. Overall, the findings from this study broaden our understanding of the prospective links of temperament traits in childhood and adulthood with illness behavior development, suggesting distinct associations from early-life illness.