Links of positive affect and stress to HbA1c: a prospective longitudinal study

Every year, 1.5 million adults in the United States are diagnosed with diabetes, with more than 90% of cases diagnosed as type 2 (US Department of Health & Human Services, 2020). People with diabetes face a complex disease management regimen that must be consistently maintained to prevent serious microvascular, neuropathic and macrovascular complications (Zheng et al., 2018). This challenging self-management regimen contributes to higher levels of daily stress, as people with diabetes work to maintain recommended diet and exercise plans, attend medical appointments, monitor glucose, and adhere to medication regimens.

Some people with diabetes struggle more with this transition to chronic disease management than others, and one factor that may influence an individual’s success in navigating this shift is positive affect (PA). PA, the experience of pleasant emotions, shows strong links to a number of positive health outcomes, including lower mortality risk and reduced risk of disease onset, with increased health behaviors serving as a major pathway through which these effects occur (Pressman & Cohen, 2005; Pressman et al., 2019). In the context of chronic illness, PA predicts reduced disease severity and slower disease progression in the domains of cancer, cardiovascular disease, and HIV/AIDS (Pressman et al., 2019). However, the role of PA in the context of diabetes remains understudied; instead research has focused on the robust detrimental effects of negative affect (NA) on disease onset, management, and outcomes (e.g., Gonzalez et al., 2007; Naicker et al., 2017; Skaff et al., 2009). While PA and NA do covary, they are conceptually and statistically distinct constructs: correlations tend to be modest, and each shows independent links to social, psychological, and physiological outcomes (Chida et al., 2008; Diener & Emmons, 1984; Steptoe et al., 2009). Thus, the emphasis on NA in diabetes gives only a partial picture as to how affect may impact adjustment to and management of this disease.

Some work has investigated PA and related constructs (e.g., resiliency, well-being) as protective factors in the context of diabetes (Robertson et al., 2012). For example, PA has been found to buffer the link between family history of diabetes and risk of developing the disease (Tsenkova et al., 2016), has been linked to increased physical activity among adults with diabetes (Spangler & Konen, 1993), and has been linked to reduced risk of all-cause mortality among people with diabetes (Moskowitz et al., 2008).

However, our understanding of the impact of PA on HbA1c, a measure of one’s average blood glucose over the prior two to three months, remains incomplete. Some cross-sectional and ecological momentary assessment studies have investigated the links between PA and HbA1c or blood glucose levels, but findings are mixed (Boylan et al., 2017; Papanas et al., 2010; Shapira et al., 2021; Skaff et al., 2009). Only a handful of studies have prospectively examined the effect of PA on HbA1c, but each of these studies is limited in some way. One study of adolescents with type 1 diabetes found PA to prospectively predicted lower HbA1c six months later, but NA was not statistically controlled (Lord et al., 2014). Thus, it is unclear whether these findings are due to high PA or the absence of NA, or if they would generalize to adults with diabetes or to people with type 2 diabetes. Another study of older women without diabetes found baseline PA to prospectively predicted lower HbA1c two years later, even after controlling for NA (Tsenkova et al., 2008). A nationally representative sample of adults in England who did not have diabetes found well-being (including constructs aside from affect, like autonomy and control) prospectively predicted lower HbA1c over a period of eight years, although PA was not directly assessed (Poole et al., 2020). While these findings are suggestive, it is unclear if results from persons without diabetes would generalize to those with diabetes. Thus, the goal of this study was to prospectively investigate the effect of PA on HbA1c among adults with type 2 diabetes.

Because it is unlikely that the role of PA in diabetes is uniform across everyone, our second study goal was to examine an important factor that might impact this relation—stress. Research has shown stress to moderate the relation between PA and health such that PA is most beneficial for individuals who report higher levels of stress (Pressman & Cohen, 2005). For example, one experiment on skin barrier disruption found PA to predict faster healing, but only among those assigned to a stress condition (Robles et al., 2009). The stress-buffering effect of PA on health has also been found in the context of immune function (Blevins et al., 2017) and mortality (Okely et al., 2017; see Pressman et al., 2019 for discussion). PA is more effective under conditions of high stress because it reduces stress appraisals and physiological stress reactivity, offers a “time out” from stress, and builds and mobilizes resources that can be used to address a stressor (Fredrickson, 2001; Pressman et al., 2019). These effects have the capacity to facilitate positive coping processes and, ultimately, better health.

To the extent that people with diabetes vary in the level of stress they experience—both in terms of general stress and stress specific to their disease—benefits of PA in relation to disease outcomes may differ across persons. Indeed, the intensity of one’s disease management regimen differs across people with type 2 diabetes, indicating one’s stress level likely varies as well. Little research has examined the interaction between stress and PA in the context of diabetes. In a notable exception, Moskowitz et al. (2008) examined whether PA buffered the effects of general subjective stress in adults with and without diabetes. They found that the association between PA and mortality was not moderated by stress among people with diabetes but was moderated by stress among older adults without diabetes. It is thus unclear if the stress-buffering effect of PA holds among people with diabetes. Further, it is unclear if diabetes-specific stress (e.g., burden of disease management, feeling that close others don’t understand this burden) is buffered by PA. The present research aims to answer these questions.

It should be noted that PA can occur across different spans of time. The present study involved secondary analysis of daily diary data, in which daily affect ratings across 14 days were averaged to form stable, dispositional measures of affect. This method avoids the recency and saliency biases associated with retrospective recall methods (Shiffman et al. 1997). Additionally, PA encompasses a range of emotions that vary in arousal level (e.g., calm versus excited). Here, we use a mid-arousal scale of general happiness as our measure of PA (Usala & Hertzog, 1989).

The primary goal of the present study was to prospectively examine whether baseline PA predicted HbA1c among adults with type 2 diabetes over a five-year follow-up period. As PA may be particularly important for disease outcomes at earlier rather than later disease stages (Pressman & Cohen, 2005), we investigated the role of PA among adults who had recently been diagnosed with type 2 diabetes. Specifically, we examined whether baseline PA predicted HbA1c at baseline (T1), six months (T2), and five years (T3), independent of baseline NA. We hypothesized that PA would predict lower HbA1c at each timepoint.

Our second goal was to assess whether stress moderated these relations. We examined whether diabetes-specific distress or overall perceived stress interacted with baseline PA to predict HbA1c at each timepoint, hypothesizing that PA would be most protective for those reporting the highest stress. Finally, because Black and White people with diabetes may differ in the glycation of hemoglobin (Bergenstal et al., 2017) and the psychosocial processes influencing health (Lincoln et al., 2003), we took advantage of the fact that we had roughly equal numbers of Black and White participants and conducted exploratory analyses to examine potential interactions between race and PA in predicting the same set of outcomes.

The present study was part of a larger study on the role of the social environment in coping with diabetes. Follow up data were collected at six months (to identify short term changes in disease adjustment and relationships) and 5 years (to identify long term changes in glycemia and behaviors). Full study procedures are documented elsewhere (Zajdel & Helgeson, 2020), and full research materials are available at osf.​io/​bz8f6. Data and analysis code are available at osf.​io/​zg9d4. All analyses were conducted using R (v4.2.0; R Core Team, 2022).

Recruitment and study procedures were approved by the Carnegie Mellon University and University of Pittsburgh Institutional Review Boards. People who had been diagnosed with type 2 diabetes in the last 5 years were recruited in southwestern Pennsylvania from September 2012 to December 2017. Participants were recruited through hospital registries, mass advertising, and health fairs. Participants had to be married or cohabitating with a romantic partner who did not have diabetes to be eligible; while both couple members took part in the larger study, the present study used data only from participants with diabetes. Because type 2 diabetes disproportionately impacts Black communities, we aimed for our sample to be roughly half Black and half White.

The COVID-19 pandemic began part way through collection of the five-year follow-up. As such, 22.8% of participants completed their five-year assessment remotely after stay-at-home orders began on March 16th, 2020.

At baseline, 207 persons with diabetes participated in the study. Because seven participants did not complete the daily diary period during which PA was measured, the baseline sample size was n = 200. Of these, we retained 192 persons at T2 and 129 at T3. Reasons for non-participation at T3 included being unable to reach the participant (n = 26), passive refusal (n = 21), declining participation (n = 8), the participant passing away (n = 7), and other reasons (n = 2). Five participants did not have HbA1c data for T3 so were dropped from analysis. Thus, the final sample size for the present study was 123. We included those who had HbA1c data at all three time-points because this facilitates direct comparison across time. That is, if the sample size differed at each wave, findings would be confounded with these changes in sample composition. Of these participants, 44.7% were female and 55.3% were male; 77.2% were married and 22.8% were cohabitating with a romantic partner. Full demographic information is presented in Table 1.

A comparison of participants lost to follow up or dropped from analyses to those who were retained in the present analyses revealed several differences in demographic variables. Those retained in the final sample were more likely to be White than Black (p = 0.02); whereas the baseline sample was 46.9% Black and 53.1% White, the final sample was 39.8% Black and 60.2% White. Those retained in the study also had lower baseline HbA1c (M = 6.89%) than those who were lost to follow-up (M = 7.63%, t = 2.81, p < 0.01). Additionally, those retained in the present analyses were more likely to have higher incomes (p = 0.006) and were marginally more likely to be married (p = 0.06) and more educated (p = 0.06). There were no differences in baseline PA, NA, perceived stress, or diabetes specific distress between the two groups.

In this study, we proposed that PA would predict lower Hba1c over time among adults with type 2 diabetes due to the health-protective effects associated with PA. Results showed that PA was cross-sectionally associated with lower HbA1c at baseline and that PA predicted lower HbA1c five years later controlling for baseline HbA1c. This pattern of results was the same regardless of participant race. This research extends prior work on the protective effects of PA from other samples (i.e., adolescents with type 1 diabetes and adults without diabetes) to adults recently diagnosed with type 2 diabetes and supports the conclusion that PA predicts positive changes in HbA1c over time.

Of note, there was no zero-order correlation between baseline PA and HbA1c at six months, and baseline PA did not prospectively predict change in HbA1c six months later. At this point it is not clear why this was the case, or indeed why HbA1c at six months was unrelated to all of our psychosocial variables. More research is needed to understand these findings. Our results suggest that the processes through which PA influences HbA1c operate on the order of years rather than months.

There are multiple mechanisms through which PA may influence blood glucose levels. One likely pathway is through health-protective behaviors. Research has found PA to predict health behaviors like diet and exercise among the general public (Kushlev et al., 2020). A growing body of literature supports similar findings in the context of chronic illness (Bassett et al., 2019; Miles et al., 2018) although findings vary based on the specific health behavior in question and the PA arousal level (e.g., calm versus excited; Jones et al., 2021). To the extent that PA helps people with diabetes better adhere to diet, exercise, and medication regimens, it may subsequently lead to lower HbA1c levels. Relationships may similarly play a key role. PA is known to foster the accrual of social resources like increased social connectedness and improved relationship quality (Ramsey & Gentzler, 2015). These social resources, in turn, have strong links to improved health outcomes across a variety of contexts (Holt-Lunstad et al., 2010; Robles et al., 2014). Of note, the present sample was composed only of individuals in long-term romantic relationships. Given the specific role of romantic relationships in health (Robles et al., 2014), this pathway may be particularly important in this sample. Future research is needed to investigate these potential mediators and compare findings to those not in romantic relationships.

Importantly, the links between PA and HbA1c were moderated by stress. At baseline, both overall perceived stress and diabetes-specific distress interacted with PA to predict HbA1c. Further, exploratory analyses revealed that T3 perceived stress interacted with baseline PA to predict T3 HbA1c. All three interactions were such that PA was most protective at higher levels of stress, consistent with the stress-buffering hypothesis of PA and health (Pressman & Cohen, 2005). However, both perceived stress interactions were not robust to sensitivity analyses. This may be the result of low power from our somewhat small sample, or may indicate that, for those with diabetes, PA is more protective when it comes to diabetes distress versus general stress. These findings align with previous work that has found PA to “undo” the negative effects of stress, for example, by speeding up physiological recovery (e.g., heart rate) after a stressful task (Fredrickson, 2013). Relatedly, as PA both facilitates the accrual of social resources and helps people access those resources (e.g., by asking for help), it may lead people to more efficiently address stressors related to diabetes and subsequently engage in better disease management behaviors. Future research should thus investigate whether coping, particularly interpersonal coping, explains the observed stress buffering effects of PA.

While we here conceptualize PA as an independent variable that leads to positive behavioral and biological changes, it should be noted that affect has bidirectional links with other important factors that may be driving the observed effects. For example, one’s physical symptoms likely covary with PA and may predict disease outcomes, or people high in extraversion, which is marked by high PA, may adjust more quickly to a diabetes diagnosis. While the present study did not measure these variables, future research should explore the extent to which PA independently predicts blood glucose beyond these related factors.

In the clinical setting, our findings suggest that assessing dispositional PA at time of diagnosis may provide valuable information to the practitioner. By identifying patients with low PA who may be at risk for higher HbA1c, psychosocial and educational resources can be targeted towards those who may need them most. Further, these findings suggest that focusing on NA or depression may not fully characterize those who are at risk for poor adjustment after a diabetes diagnosis and indicate that PA interventions may be beneficial for patients at this stage, particularly those reporting high stress.

Strengths of the present study include the use of a daily PA measure to capture mean PA, a method that avoids the recency and saliency bias associated with retrospective trait affect measures (Shiffman et al., 1997). Further, the inclusion of NA as covariate allowed us to identify the unique predictive power of PA beyond that of the shared variance between PA and NA. Our sample was also diverse in terms of income and education, and with nearly 40% of the sample Black, we were able to confirm that our findings held across both Black and White people with diabetes. Finally, the present study took place over a lengthy follow up period, allowing us to investigate long-term links of PA to blood glucose.

We also note several study limitations. First, we had relatively high attrition, with only 123 of the original 207 participants included in these analyses. Attrition was also higher for those with higher HbA1c, lower income, and for Black participants. This differential attrition may reduce the generalizability of these findings and limit our power to detect moderation by race. Additionally, baseline HbA1c was measured just prior to the daily diary period during which affect was assessed. As HbA1c captures mean blood glucose across the previous three months, the baseline findings could be interpreted as HbA1c predicting PA. In averaging daily affect, our intention was to create a stable measure of PA representative of long-term affective trends, which should thus similarly represent a participant’s affect three months prior. While this method has been recommended in the affect literature (Merz & Roesch, 2011) it is possible that this two-week span was not fully representative of dispositional affect. However, the fact that this mean PA measure predicted HbA1c 5 years later indicates that its effects were, in fact, long lasting. Future research is recommended to replicate these findings using other measures of trait PA or by aggregating over longer diary periods. Additionally, the present research used only a single PA scale capturing happiness or well-being. Given the importance of arousal level in the link of PA to health (Pressman et al., 2019; Pressmen & Cohen, 2005), future research should investigate how findings may differ for positive emotions at varying arousal levels (calm vs. excitement). Ultimately, identifying the relevant dimensions and timescales of PA in the context of diabetes could lead to standardized measures of PA for clinical use.

Several limitations pertain to the follow up period of the present study. First, there were no assessments of HbA1c between 6 months and 5 years, so we were unable to examine how the predictive power of PA and stress evolved over this period. Collection of the five-year follow up data was also interrupted by the onset of the COVID-19 pandemic. This may have introduced substantial barriers to one’s diabetes care: participants may have delayed appointments in fear of contracting COVID-19, or loss of income and/or health insurance may have impacted one’s ability to adhere to medication and diet regimens. Alternatively, the reduction in social events may have increased control over one’s diet, for example by limiting meals out and facilitating meal planning, potentially assisting in disease management. Additionally, our measure of HbA1c changed to mail-in self test kits due to stay-at-home requirements. However, our investigation into whether these changes introduced bias into our results revealed no systematic differences.

In sum, we found robust evidence that PA predicted healthier HbA1c five years later, and weaker evidence for a cross-sectional link of PA to HbA1c. We also found that PA helped buffer against the negative effects of stress, especially diabetes-specific stress, on blood glucose levels. These findings position PA, independent of NA, as a potentially indicator of who may be at lower risk for poor diabetes outcomes. Identifying people with diabetes with low dispositional PA could allow for psychosocial resources to be targeted towards those who need them most.