Psychosocial Predictors of Diabetes Risk Factors and Complications: An 11-Year Follow-up

Abstract

Objective:

The goal of this study was to use the risk and resistance framework to examine whether a set of psychosocial variables measured at age 12 in youth with type 1 diabetes would predict the emergence of diabetes risk and complication variables 11 years and 13 years later.

Method:

We interviewed youth with type 1 diabetes when they were average age 12 and followed them for 11 years until they were average age 23 and then average age 25. At age 12, we measured personality traits (unmitigated communion, unmitigated agency), relationship variables (parent relationship quality, friend support, friend conflict), indicators of psychological well-being (depressive symptoms, bulimic symptoms, self-worth), and self-care behavior. We used these psychosocial variables assessed at age 12 to predict diabetes risk factors, glycemic control, and the emergence of diabetes complications at follow-up.

Results:

Higher unmitigated agency, poor quality parent relationships, higher friend conflict, bulimic symptoms, and lower self-worth predicted one or more diabetes outcomes. When statistical controls for age 12 glycemic control were employed, unmitigated agency emerged as the most robust predictor of diabetes outcomes.

Conclusion:

Unmitigated agency, which involves an overly inflated view of the self and a cynical view others, predicted poor diabetes outcomes over an 11-year and 13-year period. The processes by which unmitigated agency could influence health are discussed.

There is a large literature on children and adolescents with diabetes that has identified a number of important psychosocial variables linked to diabetes outcomes. A cohesive and expressive family environment, supportive peer relations, resilient personality traits, and psychological well-being are linked to good self-care behavior and glycemic control, whereas a conflictual family environment, conflictual peer relations, risky personality traits, and psychological distress are linked to poor self-care behavior and glycemic control (see Helgeson, Naqvi, Van Vleet, & Zajdel, in press, for a review). The vast majority of that research is correlational and cross-sectional. There are fewer studies that have examined the implications of psychosocial variables for diabetes outcomes longitudinally.

In this paper, we adopt a risk and resistance framework (Wallander & Varni, 1998) to examine whether psychosocial variables during adolescence predict diabetes outcomes in young adulthood. The risk and resistance framework is an expansion of the stress and coping model and has been used to understand how children adapt to chronic physical disorders (Helgeson & Palladino, 2012a; Wallander & Varni, 1998). Chronic physical disorders, such as diabetes, are conceptualized as an ongoing strain. Risk factors impede adjustment, whereas resistance factors are protective factors that facilitate adjustment. Here we apply the model to understand how psychosocial variables relevant to adolescence influence diabetes outcomes in young adulthood.

Psychosocial Predictors of Diabetes Outcomes

We used the risk and resistance framework to examine four categories of psychosocial variables relevant to adolescence that could predict diabetes outcomes in young adulthood. First, we examined two risky personality traits tied to the gender-related traits literature that might impact diabetes outcomes—specifically, unmitigated agency and unmitigated communion. Second, we examined two relationship domains critical to adolescence—family and peer relations—that might be connected to those personality traits and predict diabetes outcomes. Third, we examined psychological well-being factors critical to adolescence—depressive symptoms, disturbed eating behavior, and self-worth—that might be influenced by the previously identified personality and relationship variables and also predict diabetes outcomes. Finally, we examined self-care behavior as a protective behavioral factor that could impact diabetes complications. See Figure 1 for a conceptual model of how these four sets of psychosocial factors could be related to each other and linked to diabetes outcomes.

Figure 1.

Conceptual model of the link between risk and resistance factors and diabetes outcomes

Personality.

There are two risky personality traits that have been well-studied in the field of health psychology generally, but have not received as much attention in the context of diabetes. These traits stem from the gender-related traits literature and are particularly well-suited to the study of adolescents, as adolescence has been referred to as a time of gender intensification (Hill & Lynch, 1983) when gender-related norms are salient.

One, unmitigated communion, is defined as a focus on others to the exclusion of the self (Helgeson & Fritz, 1998). It is characterized by overinvolvement in others’ problems and self-neglect. Interpersonally, unmitigated communion has been linked to being overly nurturant, intrusive, easily exploitable, and problematic relations with network members (Fritz & Helgeson, 1998; Helgeson & Fritz, 2000). In studies of adults with chronic illness, unmitigated communion has been linked to increased psychological distress, poor health behavior, and poor disease adjustment (Danoff-Burg et al., 2004; Helgeson, 2003).

Unmitigated communion also has been studied in adolescents with type 1 diabetes. In an earlier report on the present sample, unmitigated communion was linked to greater depressive symptoms, and a deterioration in self-care and glycemic control over time (Helgeson & Palladino, 2012b). Depressive symptoms partly accounted for the link of unmitigated communion to the deterioration in self-care. Unmitigated communion also was associated with problematic relationships with parents and friends and predicted an increase in relationship problems with parents and friends. Unmitigated communion has some overlap with “extreme peer orientation,” which involves sacrificing what is best for the self to gain peer acceptance (Drew, Berg, & Wiebe, 2010). In a study of adolescents with type 1 diabetes, extreme peer orientation was linked to poor glycemic control.

A second trait that has received less attention in the context of diabetes is unmitigated agency, which is defined as an extreme focus on the self to the exclusion of others (Spence, Helmreich, & Holahan, 1979). Unmitigated agency is characterized by arrogance, hostility toward others, and self-absorption. Interpersonally, unmitigated agency is related to being domineering, and cold, to antisocial behavior, and to conflictual interactions with others—including work with adolescents (Helgeson & Fritz, 2000; Spence et al., 1979). In studies of men with prostate cancer and persons with heart disease, unmitigated agency has been related to greater psychological and physical health problems (Helgeson & Fritz, 2000; Helgeson & Lepore, 2004).

Unmitigated agency has only been examined in the context of youth with type 1 diabetes once—an earlier report on the present sample (Helgeson & Palladino, 2012b). In that study, unmitigated agency was related to lower parent relationship quality, higher friend conflict, increased psychological distress and poorer self-care behavior Each of these variables partly accounted for the link of unmitigated agency to poor self-care behavior.

Thus, the previously stated links of unmitigated communion and unmitigated agency to problematic relationships, increased psychological distress, and poor self-care suggest multiple pathways by which these two traits could predict poor diabetes outcomes.

Relationships.

The relationship most often studied among children and adolescents is the family. A cohesive and expressive family environment has been linked to good diabetes outcomes, and family conflict around diabetes management has been linked to poor outcomes (see Helgeson & Palladino, 2012a, for a review). Thus, in the present study we examined whether a supportive parent relationship during adolescence is a resistance factor that predicts good diabetes outcomes in young adulthood.

Another important relationship during childhood and adolescence is the peer relationship. Yet, far fewer studies examine peer compared to family relationships. Friend support has been linked to greater psychological well-being, but links to diabetes outcomes are equivocal (Van Vleet & Helgeson, in press). By contrast, more conflictual friend relationships have been linked to poor diabetes outcomes. A longitudinal study of the present sample showed that friend conflict predicted decreases in glycemic control over four years (Helgeson, Siminerio, Escobar, & Becker, 2009). In the present study, we examine whether friend support is a risk or resistance factor and whether friend conflict is a risk factor for poor diabetes outcomes.

Psychological well-being.

Adolescence is a peak period for psychological distress (Avenevoli, Swendsen, He, Burstein, & Markangas, 2015). And, psychological distress (e.g., depressive symptoms) is associated with poor glycemic control in adolescents (Jaser, Patel, Xu, Tamborlane, & Grey, 2017). Psychological distress also has been identified as a risk factor for membership in trajectories of deteriorating glycemic control (Hilliard et al., 2013).

A mental health issue that has received substantial attention in the adolescent literature and is of particular interest in the area of diabetes is eating disorders or disturbed eating behavior. Type 1 diabetes is managed with insulin, and insulin is associated with weight gain. Girls with diabetes, in particular, learn that withholding insulin leads to a loss of weight. The rate of eating disorders and disturbed eating behavior is higher among adolescents with than without type 1 diabetes (Young et al., 2013), and both are linked with poor glycemic control (Young et al., 2013).

Finally, a third indicator of well-being is self-esteem. Self-worth has been linked with psychological well-being as well as diabetes outcomes. In one study of adolescents with type 1 diabetes, self-esteem was a component of a resilience score that was linked to reduced distress, better quality of life, and better glycemic control (Yi-Frazier et al., 2015).

Thus, we examine whether depressive symptoms and disturbed eating behavior are risk factors and self-esteem is a resistance factor in terms of future diabetes outcomes.

Behavior.

Self-care behavior is typically related to glycemic control (Helgeson, Vaughn, et al., 2018; Kristensen et al., 2012). However, the size of this relation is often modest. Here we have the opportunity to determine whether self-care behavior at age 12 is a protective factor in predicting diabetes outcomes in young adulthood.

The Present Study

We enrolled youth with type 1 diabetes when they were average age 12 and followed them for 11 years. At age 12, we measured: personality (unmitigated communion, unmitigated agency), relationships (parent relationship quality, friend support, friend conflict), psychological well-being (depressive symptoms, disturbed eating behavior, self-worth), and self-care behavior. At age 12, we had available the laboratory measure of HbA1c, demographic variables (e.g., parent social status), and background disease variables (e.g., age at diagnosis). At age 23, we invited participants to come to the laboratory to assess risk factors for diabetes complications (i.e., blood pressure, cholesterol, waist circumference, HbA1c) and early indicators of diabetes complications (i.e., nephropathy, neuropathy). With controls for demographic and disease variables, we sought to predict whether age 12 personality, relationship, psychological well-being, and self-care predicted these risk factors and complications at age 23 and 25. To the extent that personality variables predicted outcomes, we sought to determine whether we could explain those links in terms of relationship, psychological well-being, and self-care.

Method

Procedure

Recruitment.

Participants (n = 132) were recruited from Children’s Hospital of Pittsburgh when they were average age 12 (Time 1) into a longitudinal study that was focused on the transition over adolescence. Youth were eligible to participate if they were in the 5th, 6th, or 7th grade. Ages ranged from 10.70 to 14.21, with a mean of 12.08 (SD = .73). Recruitment details are provided elsewhere (see Helgeson, Snyder, Escobar, Siminerio, & Becker, 2007). At the end of the 5-year study, they were re-recruited into a second study that focused on the transition out of high school into emerging adulthood (annual assessments ages 17–19; see Helgeson et al., 2014a for re-recruitment details) and then into this final study that focused on emerging adulthood (annual assessments ages 23–25). Participants in the present report had been followed for 11 years until they were average age 23 (Time 2) and then two more years until they were average age 25 (Time 3). Demographic data on the sample at Times 1, 2 and 3 are reported in Table 1.

Table 1.

Demographic Variables of the Sample

	Time 1 (n = 132)	Time 2 (n=99)	Time 3 (n=86)
Male sex	47%	44%	42%
White race	93%	92%	91%
Non-Hispanic ethnicity	96%	96%	97%
Live with two parents	74%
Age	12.11 +/− .75 years	22.87 +/− .55 years	24.91 +/−.54
Social Status*	41.97 +/− 11.05	38% college grad	50% college grad
HbA1c %	8.18 +/− 1.27	8.83 +/− 1.68	8.75 +/− 1.72
Diabetes duration	4.92 +/− 2.97 years	15.67 +/− 3.12 years	17.61 +/− 3.18 years
Insulin Pump (vs. MDI)	26%	60%	62%

^*Measured with Hollingshead (1975) social status, which takes into consideration parent education and occupational status.

Procedure.

All procedures were approved by Carnegie Mellon and University of Pittsburgh Institutional Review Boards. Upon recruitment, the participant and a parent were met before or after an upcoming clinic appointment. Child assent and parent consent was obtained. Parents completed a brief questionnaire that provided demographic and disease background information, while children were interviewed privately to complete the psychosocial measures listed below. Children were paid $25. Parents were paid $10.

About 11 years later, participants were contacted and invited to come to our laboratory in Pittsburgh for an in-person interview and a medical assessment that included measures of the risk factors and diabetes-related complications identified below. If participants had moved out of the area, we completed interviews by phone and mailed a prescription to visit a Quest laboratory to obtain a blood draw and to provide a urine sample for the assessments described below. All participants were re-consented before any data were collected. A detailed description of the data we collected, missing data, and reasons for missing data are shown in Figure 2. Participants were paid $200.

Figure 2.

Description of recruitment, retention, and available data for all dependent measures. We used multiple imputation for missing outcome data, so all analyses are based on n = 107, the number of people that we contacted at Time 2.

¹Participants were interviewed in their homes when they lived too far from the Pittsburgh laboratory; they are more likely to have missing data on neuropathy. ²Participants were interviewed by phone when they had moved out of the area; they were missing both measures of neuropathy, the measure of waist circumference and blood pressure which are part of the risk index, but had measures of cholesterol and glycemic control if they were able to go to a Quest lab.

Psychosocial Measures

Personality.

Unmitigated communion was assessed with Helgeson’s 9-item measure (1993; Fritz & Helgeson, 1998). The wording of some items was slightly modified to be more easily understood by younger adolescents. Previous research has shown that this scale demonstrates .7 to .8 internal consistency and high test-retest reliability (Fritz & Helgeson, 1998). This scale taps placing others’ needs before one’s own and distress over concern for others. Sample items include “I often worry about others’ problems” and “Even when I don’t feel well, I always help other people” (α = .70).

Unmitigated agency was assessed with a modification of the 8-item Extended Personal Attributes Questionnaire subscale (Spence et al., 1979). The original subscale consists of adjective pairs that are rated on 5-point bi-polar scales (e.g., “not at all greedy” to “very greedy”; “not at all boastful” to “very boastful”). To facilitate comprehension for adolescents, these pairs of words were placed into sentences (e.g., I am greedy, I brag about myself a lot, I look out for myself before I look out for others, I am sarcastic). We removed one item (“I am often rude to people”) because it detracted from the reliability (α = .77).

When these two constructs were examined in an earlier report on this sample, we found that they functioned similarly to their adult counterparts (Helgeson & Palladino, 2012b). First, across five annual waves of assessment the two constructs were uncorrelated (r’s ranged from −.10 to .02, n.s.). Second, males scored higher than females on unmitigated agency, and females scored higher than males on unmitigated communion consistent with previous research. Third, unmitigated communion was positively correlated with the personality trait of communion (i.e., positive focus on others; r’s ranged from .40 to .57, p’s < .001), but distinct from communion in its ability to predict relationship and health outcomes. Unmitigated agency was positively correlated with the personality trait of agency (i.e., positive focus on the self; r’s ranged from .32 to .40, p’s < .001), but distinct in its ability to predict relationship and health outcomes. These two personality traits are rooted in the gender-related traits literature (Helgeson, 2003), which includes research on children and adolescents (Antill, Cunningham, Russell, & Thompson, 1981; Boldizar, 1991).

Parent relationship.

Using Kerr and Stattin’s (2000) parent relationship quality measure, adolescents rated the frequency of 8 items for mother and father separately from 1 (never) to 5 (very often). Items included “How often do you and your mom understand each other?” and “How often does your dad support and encourage you?” (mother: α = .79; father: α = .88). Because mother and father relationship quality were correlated (r = .36, p < .001), we averaged the two to form a parent relationship quality index.

Friend support and conflict.

We administered the 6 scales from the Berndt and Keefe (1995) friendship questionnaire: companionship, intimacy, instrumental support, self-esteem enhancement, conflict, and dominance. The first four scales reflected positive support. Thus, we standardized the four scales, summed them, and took the average to form an overall support index (α = .90). Conflict and dominance reflected negative aspects of friendship, We took their average to form an overall friend conflict index (α = .82).

Mental health.

We measured depressive symptoms with the 10-item abbreviated form of the Children’s Depression Inventory (Kovacs, 2001). Each multiple choice item has a 3-point scale. A sample item asks respondents to choose from: “I am sad once in a while,” “I am sad many times,” and “I am sad all the time” (α = .76). We measured bulimic symptoms with the 7-item subscale from the Eating Disorder Inventory (EDI; Garner, 1991; α = .75), Sample items include “I stuff myself with food” and “I eat or drink in secrecy.” Responses are made on a 5-point scale ranging from “never” to “very often.” We measured self-esteem with the 6-item global self-worth subscale from the Self-Perception Profile for Children (Harter, 1985; alpha = .75). A sample item states “Some kids are usually happy with themselves as a person, but other kids are not happy with themselves. Would you say that you are usually happy with yourself or often unhappy with yourself?” Responses are made on a 4-point scale: really unhappy, sort of unhappy, sort of happy, really happy. All of these measures are widely used with strong evidence of validity and reliability.

Self-care behavior.

We administered the 14-item Self-Care Inventory (La Greca, Swales, Klemp, & Madigan, 1988) which was updated by adding eight more contemporary items (Helgeson, Reynolds, Escobar, Siminerio, & Becker, 2007). This scale asks respondents to indicate how well they followed their physicians’ recommendations for glucose testing, insulin administration, diet, exercise, and other diabetes behaviors that have been regarded as important by the American Diabetes Association (α = .82).

Health Measures

Risk Factor Index.

Rather than analyze risk factors for future health problems individually, we computed a risk factor index that was the sum of the following five items each scored 1 for a positive finding: (1) systolic blood pressure higher than 140 and/or diastolic blood pressure higher than 90mm/Hg or taking medication for blood pressure; (2) waist circumference over 88 centimeters for women and 102 centimeters for men; (3) hemoglobin a1c (HbA1c) over 7.5%; (4) high density lipoprotein (HDL) cholesterol of 50 or lower for females and 40 or lower for males; and (5) non high density lipoprotein (nonHDL) of 130mg/ml or higher. We viewed this index as an accumulation of risk factors.

Glycemic control.

Because we had missing data on some of the individual risk factors, we opted to examine glycemic control separately in addition to including it in the risk index because we had the maximum amount of data available for this outcome. We examined glycemic control with hemoglobin A1c (assessed with DCA Vantage, Siemens) .

Neuropathy.

Peripheral neuropathy was measured with a Vibratron II device (Physitemp Instruments, Clifton, NJ) . This determines the vibratory thresholds for the hallux (great toe) on both feet using a forced choice methodology (Arezzo, 1993). We used a value of 2.40 vibration units or greater to indicate peripheral neuropathy.

Cardiac autonomic neuropathy was measured using the expiration/inspiration heart rate ratio determined in two repeated 2-minute cycles of deep breathing. The overall average ratio of heart rate during exhalation to heart rate during inhalation was calculated for both 2-minute segments. Values under 1.1 indicated cardiac autonomic neuropathy.

The two measures of neuropathy were marginally related at Time 2, X²(1) = 3.22, p = .07, and significantly related at Time 3, X²(1) = 16.19, p < .001. At Time 2, 2 individuals had neuropathy on both measures, and 15 had neuropathy on one measure. At Time 3, 7 individuals had neuropathy on both measures, and 11 had neuropathy on one measure. Because the base rate of neuropathy on either measure was fairly low, we combined the two measures and scored individuals as having neuropathy if they met the criterion for either measure (scored as 1) or neither measure (scored as 0) at Time 2 and at Time 3.

Nephropathy.

Nephropathy was measured from the albumin/creatinine ratio of a random spot urine collected in the lab. If the ratio was greater than 30, we attempted to collect an overnight urine but this was not always possible. At Time 2, we were able to obtain and verify the high value for 9 of 10 because 1 participant did not collect the overnight urine. At Time 3, we were only able to obtain and verify the high value for 4 of 12 participants. We used the value of 30 to determine nephropathy.

Overview of the Analysis

First, we examined differential attrition and missing data with respect to the measure of glycemic control and the risk factor index. We compared those who did and did not have a measure of glycemic control and those who did and did not have a risk factor index on demographic and psychosocial variables. We used the same procedure to compare participants who did and did not have complications data available at Time 2 and Time 3.

Second, we examined whether demographic or disease background variables were related to psychosocial variables, glycemic control, the risk factor index, or either complication index. These variables included participant sex, race, age, length of diabetes, age at diagnosis, parent reported social status (Hollingshead, 1975; based on education and occupational status), and household structure at age 12 (living with two parents or not). To the extent that these variables were linked to both psychosocial variables and diabetes outcomes, we statistically controlled for them in subsequent analyses.

We used multiple regression analysis to predict the risk factor index and the measure of glycemic control at Time 2 and Time 3. We used logistic regression analysis to predict the two complications variables at Time 2 and Time 3. Each psychosocial variable was examined separately, controlling for relevant background variables. We did this to reduce the number of variables involved in each model and to isolate effects of individual psychosocial predictors. We then repeated all of these analyses by controlling for Time 1 glycemic control. We used multiple imputation to account for missing outcome data at Time 2 and Time 3, so all analyses represent the number of participants we contacted at Time 2 (n =107; van Buuren & Groothuis-Oudshoorn, 2011).

We examined outcomes at both Time 2 and Time 3. We could have focused solely on Time 3 as this is when the level of risk factors and complications would be at its highest. However, we had higher attrition at Time 3 than Time 2, so Time 2 provided the maximum data. We view Time 3 as a partial confirmation of Time 2.

To the extent that we found links of personality to outcomes, we examined whether the relationship, psychological well-being, or self-care variables that were marginally or significantly related to outcomes mediated the relation of personality to outcomes. We tested mediation by examining indirect effects using Mplus with full maximum likelihood.

Results

Preliminary Analyses

When we compared respondents who did and did not have a Time 2 glycemic control value on baseline demographic and psychosocial variables, there were no group differences on any variable, with the exception of unmitigated communion, t(130) = 2.00, p < .05. Those who dropped out of the study scored lower in unmitigated communion than those who remained. We did the same comparison at Time 3, and again found only one difference on the same variable, unmitigated communion, t(130) = 2.33, p < .05.

When we compared respondents who did and did not have a Time 2 risk factor index on baseline demographic and psychosocial variables, those who remained in the study and had a risk factor index were higher in social status (M = 43.68, SD = 10.68) than those who dropped out (M = 38.53, SD = 11.09), t(130) = 2.58, p < .05. The same comparison at Time 3 revealed no group differences, with the exception of the same difference in unmitigated communion described above, t(130) = 2.82, p < .01. There were no differences between participants who did and did not have Time 2 or Time 3 complications data available on baseline demographic, disease, or psychosocial variables.

We also examined whether psychosocial variables were related to demographic variables. Older age was related to poorer self-care behavior (r = −.27, p < .001), as was a longer time since diagnosis (r = −.23, p < .01). Household structure was related to only one psychosocial variable, parent relationship quality: children from two-parent families reported better parent relationships (M = 4.11, SD = .53) than children from non two-parent families (M = 3.87, SD = .57). Child sex was related to several psychosocial variables, all in the directions to be expected. Compared to males, females reported more support from friends (M = 3.98, SD = .51 vs. M = 3.37, SD = .68), more depressive symptoms (M = 1.19, SD = .15 vs. M = 1.11, SD = .15), lower self-worth (M = 3.48, SD = .45 vs. M = 3.65, SD = .34), lower unmitigated agency (M = 2.25; SD = .68 vs. M = 2.52, SD = .78), and higher unmitigated communion (M = 2.91, SD = .67 vs. M = 2.69, SD = .63). There were no relations of parent social status or child race or ethnicity to psychosocial variables.

In testing the relations of psychosocial variables to health outcomes, we examined whether there were any demographic or disease variables that needed to be statistically controlled. Of the variables shown in Table 1, parent social status at age 12, household structure at age 12, and participant race (white, nonwhite) were individually related to one or more of the three outcomes at the two assessments. However, when all three variables were entered into regression equations to predict the three outcomes, race was consistently not significant. Thus, we controlled for age 12 parent social status and household structure in all analyses.

The correlation matrix of all psychosocial predictor variables as well as their means and standard deviations are shown in Supp Table 1. The frequency of risk factor index scores and the percentage of people with each complication are shown in Supp Table 2.

Prediction of Risk Factors

The results from the multiple regression analyses are shown in Table 2. Shaded values are ones that remained significant when Time 1 HbA1c was statistically controlled.

Table 2.

Prediction of Risk Factors (Unstandardized Betas and Confidence Intervals)

	T2 HbA1c		T3 HbA1c		T2 Risk		T3 Risk
n	(99)		(86)		(88)		(69)
T1 Variables
Parent Relationship	−.31	(−.94; .31)	.31	(−.32; .93)	−.52*	(−1.00; −.04)	−.47+	(−1.02; .07)
Friend Support	−.21	(−.71; .29)	.17	(−.33; .67)	.04	(−.34; .42)	.35	(−.08; .79)
Friend Conflict	.35	(−.17; .86)	.32	(−.19; .82)	.38+	(−.00; .76)	.53*	(.06; .99)
Depressive symptoms	−.47	(−2.21; 1.28)	−.25	(−1.99; 1.48)	.31	(−1.08; 1.69)	1.18	(−.33; 2.69)
Bulimic symptoms	.41	(−.25; 1.07)	.67+	(−.03; 1.36)	.45+	(−.06; .96)	.64*	(.03; 1.25)
Self-Worth	−.16	(−1.03; .72)	−.13	(−1.06; .79)	−.45	(−1.14; .25)	−.27	(−1.10; .55)
Unmitigated Agency	.50*	(.07; .94)	.60*	(.11; 1.08)	.24	(−.10; .59)	.53**	(.14; .92)
Unmitigated Communion	−.37	(−.90; .15)	.23	(−.29; .76)	.25	(−.16; .97)	.33	(−.13; .80)
Self-Care behavior	−.13	(−.98; .73)	.26	(−.61; 1.13)	−.02	(−.69; .64)	−.01	(−.78; .80)

Note:

⁺p<.10

^*p<.05

^**p<.01

^***p<.001

these analyses control for parent social status and household structure; shaded values indicate those that are significant when Time 1 HbA1c is statistically controlled.

Glycemic control.

The only variable that significant predicted higher levels of glycemic control at Time 2 was unmitigated agency. At Time 3, unmitigated agency predicted poorer glycemic control, and bulimic symptoms was a marginal predictor of poor control. When statistical controls for Time 1 HbA1c were included, unmitigated agency remained a significant predictor at Time 2 and Time 3.

Risk Factor index.

Parent relationship quality individually predicted the Time 2 risk factor index, such that poorer quality relationship with parents predicted a higher risk factor index at Time 2. When statistical controls were included for Time 1 HbA1c, parent relationship quality remained a significant predictor. Higher friend conflict and bulimic symptoms were marginal predictors of a higher risk index.

At Time 3, friend conflict, bulimic symptoms, and unmitigated agency predicted a higher risk factor index. These remained significant with controls for Time 1 HbA1c. Lower parent relationship quality was a marginally significant predictor of a higher risk index.

Prediction of Complications

The results of the logistic regression analysis are shown in Table 3. Shaded values are ones that remained significant when controls for Time 1 HbA1c were employed.

Table 3.

Logistic Regression: Prediction of Complications (Odds Ratios and Confidence Intervals)

	T2 Neuropathy	T3 Neuropathy	T2 Nephropathy	T3 Nephropathy
n	(88)	(71)	(97)	(84)
T1 Variables	Odds Ratio	Odds Ratio	Odds Ratio	Odds Ratio
Parent Relationship	.50 (.18; 1.40)	.58 (.22; 1.52)	1.60 (.32; 8.01)	.93 (.10; 1.52)
Friend Support	.92 (.41; 2.10)	1.23 (.54; 2.84)	3.42 (.66; 17.6)	.90 (.30; 2.68)
Friend Conflict	1.88+ (.89; 3.99)	2.14+ (.97; 4.75)	1.64 (.56; 4.80)	1.78 (.73; 4.30)
Depressive symptoms	8.26+ (.70; 97.29)	1.64 (.15; 18.44)	9.10 (.35; 238.82)	1.23 (.07; 22.34)
Bulimic symptoms	2.55+ (.94; 6.91)	2.73+ (.94; 7.94)	2.23 (.61; 8.19)	2.25 (.66; 7.60)
Self-Worth	.21* (1.10; 4.2)	.49 (.12; 1.97)	.64 (.07; 5.81)	.85 (.14; 5.03)
Unmitigated Agency	2.16* (1.10; 4.25)	3.58** (1.57; 8.18)	1.70 (.62; 4.63)	2.72* (1.14; 6.52)
Unmitigated Communion	1.24 (.54; 2.85)	1.49 (.67; 3.34)	.71 (.20; 2.51)	1.11 (.35; 3.52)
Self-Care behavior	−.42 (.10; 1.81)	.90 (.23; 3.48)	1.26 (.15; 10.44)	.22 (.03; 1.72)

Note:

⁺p<.10

^*p<.05

^**p<.01

^***p<.001

these analyses control for parent social status and household structure; shaded values indicate those that are significant when Time 1 HbA1c is statistically controlled.

Neuropathy.

At Time 2, lower self-worth and higher unmitigated agency emerged as significant predictors of neuropathy. Higher friend conflict, more depressive symptoms, and greater bulimic symptoms emerged as marginal predictors of neuropathy. When statistical controls were included for Time 1 HbA1c, only self-worth remained significant. At Time 3, unmitigated agency predicted neuropathy, and remained significant with statistical controls for Time 1 HbA1c. Friend conflict and bulimic symptoms were marginally significant predictors of neuropathy.

Nephropathy.

At Time 2, no variables were significant predictors of nephropathy. At Time 3, unmitigated agency emerged as a significant predictor. When statistical controls for Time 1 HbA1c were included, there were no significant predictors of nephropathy.

Mediation

Because unmitigated agency predicted a number of outcomes, we could examine whether any of the relationship or psychological well-being variables accounted for this relation. Bulimic symptoms did not mediate the relation of unmitigated agency to Time 3 HbA1c. The relation of unmitigated agency to the Time 3 risk index was not significantly mediated by parent relationship quality, friend conflict, or bulimic symptoms. However, the indirect effects for friend conflict (estimate = .09; CI = −.01 to .20; p = .09) and bulimic symptoms were marginally significant (estimate = .09; CI = .01 to .22; p = .09). Neither friend conflict, depressive symptoms, bulimic symptoms, or self-worth mediated the relation of unmitigated agency to Time 2 neuropathy. Neither friend conflict nor bulimic symptoms mediated the relation of unmitigated agency to Time 3 neuropathy.¹

Discussion

This longitudinal study afforded us the unique opportunity to link psychosocial variables during adolescence to diabetes outcomes in early adulthood. We used the risk and resistance framework to examine a set of psychosocial variables that are especially critical during adolescence for well-being and sought to determine if they would have implications for health a decade later among young adults. We examined risky personality traits, risk and protective relationship variables, risk and protective psychological well-being variables, and protective self-care behavior. We found some evidence for links of both risk and protective factors to diabetes outcomes. Taken collectively, risk factors that predicted one or more diabetes outcomes 11 or 13 years later included unmitigated agency, friend conflict, and bulimic symptoms. The two protective factors were a high quality parental relationship and high self-worth. By contrast, depressive symptoms, unmitigated communion, and self-care behavior did not predict outcomes.

The psychosocial variable that revealed the most robust relation to diabetes outcomes over a decade later was unmitigated agency. Even with controls for baseline levels of glycemic control, unmitigated agency predicted a deterioration in glycemic control at both assessments and the later measures of the risk factor index and neuropathy. Unmitigated agency is a personality trait characterized by an overinflated view of the self and a cynical view of others. Although unmitigated agency has been linked to poor adjustment to heart disease (Helgeson, 1993) and prostate cancer (Helgeson & Lepore, 2004), this construct has rarely been examined in diabetes and less often in adolescents.

Unfortunately, we were not able to discern from these data why unmitigated agency predicted poor diabetes outcomes. The effects of unmitigated agency could have been observed at any point during the 11 or 13 years and may have, in fact, been cumulative. Because the mediators were assessed at the same time and are likely to have changed, our ability to detect mediation was limited.

However, we do have information about unmitigated agency from an earlier report on this sample that may help to understand these findings. When the participants in this sample were adolescents, we found that unmitigated agency assessed over five annual assessments was linked to poorer quality relationships with parents, increased friend conflict, and increased anger (Helgeson & Palladino, 2012b). That study showed that the relation of unmitigated agency to poor self-care was partly explained by poorer parent relationship quality, higher friend conflict, and elevated levels of anger. Thus, unmitigated agency is clearly characterized by interpersonal difficulties that could reduce one’s capacity to garner support from friends and family as well as health care professionals. In addition, the elevated levels of anger and hostility characteristic of unmitigated agency could interfere with taking care of the self and being responsive to physician instructions. Previous research has linked unmitigated agency to poor health behavior, including the failure to adhere to physician instructions (Helgeson, 2003) and to antisocial behavior (Helgeson & Fritz, 2000). Unmitigated agency is also connected to hostility, and there is a large literature linking hostility to poor cardiovascular outcomes (see Siegman & Smith, 1994, for a review). Thus, there could be a pathophysiological link between this trait and heightened reactivity or inflammation that accounts for links to poor diabetes outcomes. Future research should aim to identify behavioral and physiological correlates of unmitigated agency that could account for these links to poor health.

A second variable that was connected to many of the outcomes is bulimic symptoms, although most of the relations were marginally significant and only one connection remained significant when controls for baseline glycemic control were included. Bulimic symptoms peak in late adolescence and early adulthood, but were assessed during early adolescence here. One wonders if early indicators of eating disturbances have long-term consequences or signify persistence of problems that continue to have health costs, which is especially the case in diabetes (Rydall, Rodin, Olmsted, Devenyi, & Daneman, 1997).

Given that psychosocial variables were assessed at age 12 when relationships and psychological well-being are changing, it is impressive that any of these variables are connected to diabetes outcomes over such a long follow-up. We were not surprised that parent and friend relationship variables rarely predicted outcomes, as the nature of both relationships are likely to have undergone many changes over middle and later adolescence, as well as early adulthood. Psychological well-being also can be volatile during this time period. Because unmitigated agency is considered to be a personality trait, it may be more stable over time, which could have enhanced its predictive power. One reason that we may not have found evidence for mediation in this study is that the mediators are more likely to have changed over the intervening years than the personality trait variables, which are typically presumed to be more stable over time.

Before concluding, we acknowledge a number of study limitations. First, we had significant attrition and variable numbers of participants for each outcome over 11 and 13 years. Some participants were unreachable, and others had moved out of the area. Even among those we reached, we were not always able to obtain the risk and complications measures. However, it is noteworthy that we had little differential attrition with respect to demographic or disease variables. Second, the sample overall was small and homogeneous with respect to race and ethnicity, detracting from the generalizability of these findings. However, the sample was normative in the sense that scores on the psychosocial variables examined in this study did not differ from an age, sex, and race matched comparison group of healthy adolescents (Helgeson & Palladino, 2012b; Helgeson, Reynolds, et al., 2007; Helgeson, Snyder, et al., 2007). Third, we examined personality variables that have been studied far more among adults than children. Fourth, we conducted a large number of analyses that have the potential to capitalize on chance findings. For this reason, we have emphasized results that are consistent across outcomes rather than isolated effects for single variables. Finally, there are other interpersonal and behavioral variables that could explain the links of unmitigated agency to diabetes risk and complication variables that we failed to capture. There are other ways to measure self-care behavior, such as objective measures of blood glucose checking or clinic attendance, that we did not assess. There also may be other sources of interpersonal difficulties, such as relationships with physicians, parents, or friends at other points during the 11-year-period, that affected diabetes outcomes. Youth characterized by unmitigated agency may display a general pattern of antisocial behavior that interferes with taking care of diabetes as well as taking advantage of interpersonal resources that could benefit diabetes outcomes.

Despite these limitations, we were able to follow the vast majority of adolescents with diabetes for 11 years and found that some psychosocial variables predicted the development of poor glycemic control, increased cardiovascular risk factors, and early diabetes complications—the most notable of which is unmitigated agency. Future work in this area should find ways to identify these at-risk children during adolescence.