Ecological Momentary Assessment of Positive and Negative Affect and Associations with Blood Glucose in Teens with Type 1 Diabetes

Abstract

Background:

Affect (i.e., emotions) can be associated with diabetes self-care and ambient glucose in teens with type 1 diabetes (T1D). We used momentary sampling to examine associations of daily affectwithblood glucose (BG) monitoring,BG levels,and BG variability in teens with T1D.

Method:

Over 2 weeks, 32 teens reported positive and negative affect (Positive and Negative Affect Scale) and BG levels on handheld computers 4x/day, coordinated with planned daily BG checks. BG values were classified as: in-range (70-180 mg/dL); low (<70 mg/dL); severe low (<54 mg/dL); high (>180 mg/dL); severe high (>250 mg/dL). Daily BG variability was derived from BG coefficient of variation (BGCV). To determine associations of positive and negative affect with BG checks, BG levels, and BGCV, separate generalized estimating equations were performed, adjusting for demographic and diabetes-related variables, for the overall sample and stratified by HbA1c (≤8%, >8%).

Results:

Teens (44% male, ages 14-18, 63% pump-treated, HbA1c 8.8 ± 1.4%) reported 51% in-range, 6% low (2% severe low), and 44% high (19% severe high) BG. In teens with HbA1c ≤8%, positive affect was associated with in-range BG (OR = 1.08, 95% CI = 1.04-1.13, P = .0002), reduced odds of very low glucose (OR = 0.35, 95% CI = 0.16-0.74, P = .006), and less daily BGCV (β = −0.9; 95% CI = −1.6, −0.2; P = .01). In teens with HbA1c >8%, negative affect was associated with less likelihood of checking BG (OR = 0.75, 95% CI = 0.64-0.87, P = .0003).

Conclusions:

Our findings shed light on individual differences in metabolic reactivity based on glycemic levels and the importance of incorporating affect into automated insulin delivery systems.

Introduction

Suboptimal glycemic control is characterized by elevated HbA1c and limited in-range glucose (70-180 mg/dL), ¹ common occurrences in teens with type 1 diabetes. ² Furthermore, glucose variability can impose a threat to achieving target glycemic control. Stress and affect (ie, emotions) can potentially yield hyperglycemia and hypoglycemia as well as glucose variability.^3-5 Most likely, the nature of the relationship between affect and glucose is bi-directional. On the one hand, affect and stress, in some individuals, can yield hypoglycemia, potentially related to the decrease in food intake. Alternatively, hormonally mediated stress responses (counter regulation) can also increase glucose. Further, negative affect can impact one’s ability to engage in diabetes self-care, leading to suboptimal glycemic control. One study of 36 adults with type 1 diabetes showed that over the course of 48 hours, hyperglycemia measured via continuous glucose monitoring (CGM) was associated for the most part with negative affect while positive affect was associated with less hyperglycemia. ⁴ This illustrates how complex the relationship is between affect and glucose.

Conflicting findings have emerged from the literature on the relationship between affect and glucose and glucose variability. ⁶ For example, a review paper identified differences across eight studies in the measurements of affect, the variable length of study time, and different technologies used to measure glucose (CGM vs. blood glucose [BG] meters). ⁶ Another potential explanation for the various observations may reflect individual differences in metabolic reactivity to stress.^3,7 Riazi et al ⁷ reported substantial variations between reported negative experiences and emotional response to glucose level. There may also be psychosocial factors (eg, treatment adherence, social support) that may contribute to variations in stress-reactivity in persons with diabetes.

Ecological momentary assessment (EMA) is an innovative methodology that repeatedly assesses participants’ behavior and affective responses in real time and in their natural environment. This design has the potential to bypass retrospective recall biases and can measure varying behaviors and affective states throughout the day. Pairing momentary affect with glucose can help elucidate this complex relationship and may help to guide future automated insulin delivery systems that will account for the changing insulin requirements that may occur during stress-related daily events. This study used ecological momentary sampling techniques to collect “real time” data regarding affect, glucose monitoring using finger sticks, and BG levels. We examined the associations between momentary affect with BG checks, BG levels, and glycemic variability in a sample of teens with T1D. Given that youth with different HbA1c profiles tend to have different levels of glucose variability that may inherently be the bases of differences in metabolic reactivity, we also explored if such associations varied according to overall level of glycemic control based on HbA1c.

Our specific aims for this study were (1) to determine the associations between momentary positive and negative affect on adherence to glucose checks; (2) to determine the associations between momentary positive and negative affect on BG levels; and (3) to examine the associations between positive and negative affect on glucose variability within 24-hour periods. We hypothesized that teens with T1D reporting more negative affect would be less likely to perform glucose monitoring. Further, we hypothesized that negative affect would be associated with extreme glucose levels, whether hyperglycemic or hypoglycemic, and positive affect would be associated with more glucose levels in-range.

Methods

Participants

Teens with T1D who received care in an urban pediatric diabetes clinic were recruited to participate in an observational study using EMA. Eligibility criteria included age 14 to 18 years, a diagnosis of T1D for at least 1 year, insulin pump therapy or multiple daily injections (at least 3 per day), and fluency in English. Individuals were ineligible for the study if they had developmental disabilities, cognitive disorders, or a major psychiatric illness that would interfere with study participation. The study was approved by the institutional review board. Prior to completing any study procedures, parents and teens aged 18 provided written informed consent and teens aged 14 to 17 provided written assent.

Procedures

This was a secondary data analysis of a 2-week observational EMA study. ⁸ Teens used handheld devices (Palm Tungsten E2 personal digital assistant; CERTAS software, PICS, Inc., Reston, VA) to complete momentary assessments about positive and negative affect 4 times per day upon prompting at times chosen by participants to coordinate with their expected BG checks. Thirty minutes after each momentary assessment, participants received an adherence query asking if they had checked their BG level (yes, no) and asking them to report their BG level based upon their fingerstick. The positive and negative affect questions were omitted at one of the 4 scheduled times each day; the omitted time varied by day. Participants could choose to receive queries at different times each day to accommodate their schedules. Queries took about 2 to 3 minutes to complete. Socio-demographic and diabetes-related data were obtained from the medical record and self-reported assessments. HbA1c using the Roche Cobas Integra method (reference range 4.0%-6.0%) assessed glycemic control.

Positive and Negative Affect

Participants rated 6 positive affect states (feeling interested, strong, proud, alert, inspired, determined) and 6 negative affect states (feeling distressed, upset, guilty, scared, hostile, irritable) on a 5-point Likert-scale (1 = not at all, 5 = extremely). ⁹ The items were adapted from the Positive and Negative Affect Schedule and used in previous studies.^8-10 The positive and negative affect scores were calculated from the sum of the responses (range for each subscale: 6-30; Cronbach’s α = 0.91 and α = 0.82, respectively), adjusting for missing values by normalizing to a total score of 6 to 30 for each of the 3 individual daily queries. The positive and negative affect mean scores were poorly correlated (r = −0.15, P = .42).

Data Analyses

The current analyses only included data in which affect responses could be paired temporarily with adherence queries and glucose reports, resulting in the inclusion of 32 of the original 36 participants. ⁸ Socio-demographic and diabetes-related characteristics of those who were excluded from the present analyses did not differ from the current sample.

Statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, NC) statistical software. Descriptive statistics were used to characterize the sample and momentary responses. For the first aim, a logistic generalized estimating equation model was applied to determine the associations between momentary positive and negative affect on adherence to glucose checks. In these analyses, we only included times when participants were supposed to check their glucose based on their scheduled BG checks. Separate models for positive and negative affect predicting glucose monitoring, adjusted for age, sex, diabetes duration, pump therapy, and HbA1c, were fit for the entire sample and then stratified according to level of glycemic control (HbA1c ≤8 and >8%).

For the second aim, a logistic generalized estimating equation model was applied to determine the associations between momentary positive and negative affect on BG levels. We classified the BG values into the following categories that correspond with clinical ranges: (1) in-range 70-180 mg/dL; (2) low <70 mg/dL; (3) very low <54 mg/dL; (4) high >180 mg/dL; (5) very high >250 mg/dL. Separate models for positive and negative affect predicting each of the BG categories, adjusted for age, sex, diabetes duration, and pump therapy, were fit for the entire sample and then stratified according to level of glycemic control (HbA1c ≤8 and >8%).

For the third aim, a linear generalized estimating equation model was applied to examine the associations between positive and negative affect on glucose variability within 24-hour periods. To determine the daily glucose variability, we calculated the blood glucose coefficient of variation (BGCV) per day: BGCV = (SD/Mean)*100. These calculations yielded a daily BGCV. Similarly, each day’s positive and negative affect scores yielded a daily positive and negative affect mean score. Separate models for positive and negative affect predicting BGCV, adjusted for age, sex, diabetes duration, and pump therapy, were fit for the entire sample and then stratified according to level of glycemic control (HbA1c ≤8 and >8%). In order to analyze the coefficient of variability, we only included 24-hour blocks in which participants reported ≥3 BG results, resulting in 265 (59%) of the 24-hour blocks.

For all analyses, a P value of <.01 was deemed statistically significant in order to reduce the likelihood of a type 1 error. However, data are reported for those associations with P values of .01 to <.05 for exploratory purposes.

Results

Study Sample

Demographic characteristics of the 32 teens with T1D are shown in Table 1. Briefly, the sample had a mean age of 16.6 ± 1.4 years and duration of 8.8 ± 4.2 years; overall frequency of BG monitoring was 3.9 ± 1.4 times per day and mean HbA1c was 8.8% ± 1.4%. The majority received insulin pump therapy; none were using CGM so the sample relied exclusively on BG results. Characteristics did not differ by HbA1c strata. In brief, the median 2-week EMA response rate was 72%. Overall, participants completed a total of 1193 BG reports and 963 affect reports, resulting in 799 pairs of completed affect and BG reports. Teens reported a total of 961 BG values, of which 51% were in-range (70-180 mg/dL); 5% were low (<70 mg/dL), with 2% of all BGs being very low (<54 mg/dL). The remaining 44% were above range (>180 mg/dL), with 19% of all BGs being very high (>250 mg/dL). BGCV was 47.3% ± 11.4%.

Table 1.

Participant Characteristics.

	Overall (N = 32)	HbA1c groups		P
		≤8% (n = 9)	>8% (n = 23)
Age (years)	16.6 ± 1.4	16.4 ± 1.5	16.7 ± 1.4	.69
Body Mass Index (z-score)	0.8 ± 0.6	0.9 ± 0.7	0.8 ± 0.6	.92
Sex (% female)	56	56	57	.96
T1D duration (years)	8.8 ± 4.2	7.5 ± 4.8	9.4 ± 4.0	.27
Pump therapy (%)	63	44	70	.19
Daily insulin dose (u/kg)*	0.9 ± 0.2	1.0 ± 0.2	0.9 ± 0.2	.58
BG monitoring frequency (times/day)	3.9 ± 1.4	3.6 ± 1.3	4.0 ± 1.5	.40
HbA1c (%)	8.8 ± 1.4	7.3 ± 0.4	9.4 ± 1.1	<.0001

^*Daily insulin dose is available for n = 31.

Positive and Negative Affect Predicting BG Checks Over 2 Weeks

In the entire sample, positive affect was not associated with report of BG checking (Table 2a) while endorsement of negative affect was related to a 17% reduced likelihood of reporting a BG check (OR = 0.83, 95% CI: 0.74-0.94; P = .003) (Table 2b). Given the potential for differences in BG monitoring between those with more optimal glycemic control, we examined the association of positive and negative affect on BG checks in teens with HbA1c ≤8% and >8%. Indeed, positive affect was associated with a trend towards being more likely to check BG for those with HbA1c ≤8% (Table 2a). Notably, in teens with HbA1c >8%, negative affect reduced the likelihood of BG checking by 25% (OR = 0.75, 95% CI: 0.64-0.87, P = .0003) (Table 2b). None of the demographic or diabetes-specific variables were related to BG checking.

Table 2. (a).

Positive Affect Predicting BG Checks and Glucose Levels.

	Overall			HbA1c ≤8%			HbA1c >8%
	OR	95% CI	P	OR	95% CI	P	OR	95% CI	P
BG checks	1.01	0.94-1.08	.66	1.18	1.03-1.35	.02	0.98	0.89-1.07	.69
In range	0.99	0.98-1.01	.89	1.08	1.04-1.13	<.0001	1.00	0.98-1.02	.69
High	1.00	0.98-1.02	.41	0.95	0.89-1.02	.21	1.00	0.97-1.02	.97
Very high	0.99	0.95-1.02	.69	0.99	0.92-1.07	.96	0.97	0.93-1.02	.35
Low	0.94	0.88-1.02	.12	0.86	0.70-1.05	.14	0.97	0.91-1.02	.34
Very low	0.95	0.81-1.11	.55	0.35	0.16-0.74	.006	0.95	0.86-1.05	.37

In-range 70-180 mg/dL, high >180 mg/dL, very high >250 mg/dL, low <70 mg/dL, very low <54 mg/dL.

Bold P value of <.01 and italic P values of .01 to <.05.

(b) Negative Affect Predicting BG Checks and Glucose Levels.

	Overall			HbA1c ≤8%			HbA1c >8%
	OR	95% CI	P	OR	95% CI	P	OR	95% CI	P
BG checks	0.83	0.74- 0.94	.003	0.93	0.77-1.14	.54	0.75	0.64- 0.87	.0003
In-range	0.97	0.94-1.00	.14	0.93	0.87-1.00	.05	0.99	0.95-1.03	.64
High	1.03	0.99-1.08	.15	1.07	1.00-1.15	.74	1.07	1.00-1.15	.03
Very high	1.06	1.03-1.10	.0006	1.12	1.04-1.19	.0006	1.06	1.01-1.11	.01
Low	0.96	0.87-1.05	.41	0.97	0.86-1.09	.64	1.00	0.91-1.10	.88
Very low	0.82	0.69-0.98	.03	0.74	0.57-1.98	.04	0.89	0.75-1.05	.17

In-range 70-180 mg/dL, high >180 mg/dL, very high >250 mg/dL, low <70 mg/dL, very low <54 mg/dL.

Bold P value of <.01 and italic P values of .01 to <.05.

Positive and Negative Affect Predicting BG Levels Over 2 Weeks

Positive affect

In the entire sample, positive affect was not associated with glucose level. However, among the teens with HbA1c ≤8%, positive affect was associated with greater in-range glucose levels (OR = 1.08, 95% CI: 1.04-1.13, P = <.0001) (Table 2a). In this model, there was a trend toward significance for younger teens being more likely to report glucose levels in-range. Positive affect was also related to reduced odds of having very low glucose (OR = 0.35, 95% CI: 0.16-0.74, P = .006) in the teens with HbA1c ≤8%. Shorter diabetes duration (P = .004) and being male (P < .0001) were associated with lower odds of very low glucose levels. Using injection-based therapy trended toward significance for lower odds of very low glucose levels.

Negative affect

In the entire sample, negative affect was related to increased odds of reporting very high glucose levels >250 mg/dL (OR 1.06, CI 1.03-1.10, P = .0006) and a trend for a lower odds of reporting very low glucose levels <54 mg/dL (Table 2b). The odds for glucose levels in the very high category was especially higher for those teens with HbA1c ≤8% (OR = 1.12, 95% CI: 1.04-1.19, P = .0006) while there was only a trend for negative affect to increase the odds of very high levels in those with HbA1c >8%. Pump use was also associated with less likelihood of having very high BG levels. Finally, there was a trend for negative affect to be associated with teens being less likely to report very low glucose levels in teens with an HbA1c ≤8%.

Positive and Negative Affect Predicting BGCV Within 24-hour Intervals

Momentary affect was assessed against daily glucose variability reported as BGCV. Positive and negative affect were not associated with BGCV for the overall sample but associations were evident in those with lower HbA1c levels. Teens with HbA1c ≤8% had a trend towards lower daily BGCV when they reported positive daily affect. This model also showed that teens who used pump therapy had significantly lower daily BGCV (β = −6.3; 95% CI: −10.2, −2.3; P = .002). Negative affect was not associated with BGCV.

Discussion

Use of EMA provided for the collection of affect, glucose monitoring adherence, and glucose values during daily living in a sample of teens with T1D. This exploratory study identified various associations between ambient affect and glucose levels and glucose variability, supporting the likely presence of important individual responses between affect and glucose as previously reported.^3,7 There are 4 main findings that we report in this study.

First, it is notable that associations between affective states and glucose varied based on glycemic levels, with most associations between affect and glucose levels derived for teens with HbA1c ≤8%. For teens with HbA1c ≤8%, positive affect increased the likelihood of in-range levels and reduced the odds of very low glucose. On the other hand, negative affect was associated with increase of very high glucose for the full sample, but was especially driven by those with HbA1c ≤8%. For teens with HbA1c >8%, negative affect was significantly associated with reduced odds of checking BG. Although this association was significant in the full sample, it was mainly driven by teens with HbA1c >8%. These observations suggest that teens with less well controlled diabetes likely have many other factors beyond affect that are associated with their suboptimal control, likely related to inadequate diabetes self-care that result in insulin excursions and glycemic variability. It is possible that these individuals do not demonstrate the same metabolic reactivity that teens with better glycemic control do.

Second, this study contributes to the literature on affect and adherence to BG monitoring. As expected, at times when teens reported negative affect, they were half as likely to check their BG; this association seemed to be driven from teens with HbA1c >8%. This finding is consistent with previous reports in teens with T1D that showed that more depressive symptoms were associated with less optimal diabetes self-care. ¹¹ This suggests that in teens with elevated glycaemia, reduced diabetes self-care behaviors are likely related to negative mood, and provides an opportunity for interventions focused on recognizing negative mood to help teens maintain their diabetes self-care behaviors at these times.

Third, negative and positive affect were associated with clinically defined BG levels. We found that while negative affect was not associated with glucose values in-range or below range, it was associated with glucose values in the very high range. This finding has been previously observed in 206 adults with type 2 diabetes who displayed an association between negative affect and next-morning hyperglycemia over the course of 21 days. ¹² Negative affect likely induces greater counter-regulation which leads to higher glucoses. In our sample, as anticipated, we found that positive affect was associated with in-range glucose levels. We were somewhat surprised from the finding that positive affect was associated with less very low BG in teens with HbA1c ≤8%. This finding is in contrast to Wagner et al ⁵ who reported on adults with type 2 diabetes using blinded CGM and reported affect twice a day over a 7 day study; they described an association between positive affect and hypoglycemia. There were substantial differences between our sample and theirs such that their sample included only adults with type 2 diabetes of whom only 57% were insulin users. ⁵ A recent review paper also found variations between samples of persons with type 1 and type 2 diabetes. ⁶

Fourth, on a momentary basis, affect was not associated with BG variability; although we observed a trend toward significance in teens with HbA1c ≤8%, where more positive affect was associated with less BG variability within 24-hour intervals. This finding differs from one study that showed that daily stress, as measured by a single daily question, was associated with glucose variability obtained from CGM data over a week period. ³ More research is needed to clarify the associations between affect and BG variability. It would be interesting to examine these relationships in closed loop systems since these systems provide automated assistance that aims to produce less glucose variability.

Another important factor to consider is the timing of stress and food intake. Prior studies show that while there is a significant delay in postprandial glucose recovery after food intake during stress, this effect was not present in the fasting state.^13,14 In addition, negative affect may be tied to food intake, as one of the known depressive symptoms is alterations in appetite. ¹⁵ Pump data can allow for exploration of these relationships; future studies can account for the impact of food intake and its association with affect and glucose.

This study has several limitations. While EMA study designs have the potential to reduce recall bias, teens self-reported their BG levels following a prompt and so we did not have objective data to report. Further, this study was performed when CGM use was limited in this age group ¹⁶ due to availability of earlier generations of CGM devices with poorer performance than current devices coupled with the ongoing need for BG monitoring for management as non-adjunctive CGM use was not yet available. The lack of CGM data contributed to a few limitations. First, our findings pointed to an association between greater negative affect and performing less BG checks, thereby limiting our ability to see what glucose values are at times of negative affect. Our data is likely an underestimation of the impact of negative affect on glucose. Second, we used relatively few daily BG levels to calculate the BGCV. Future studies can address these limitations by using CGM data to investigate the relationship between affect and glucose as well as glucose variability. In addition, in all GEE models we paired the reports between affect and BG, which led to a smaller number of observations available. Additionally, our sample size was comparable to other EMA studies but like other EMA studies, this relatively small sample size may impact the generalizability of our findings to a larger sample of teens with T1D. Further, the relatively small sample size may be limiting the statistical power to identify small associations between affect and glucose control. Last, findings from the primary outcome paper showed that the EMA signal response possibly decreased teens response rate ⁸ and this may have potentially contributed to the observed associations.

Conclusions

There appears to be clinical benefit to incorporate affect into future design of the artificial pancreas, given the association of negative affect with higher glucose levels. Our preliminary findings highlight the impact of momentary affect and diabetes control (ie, glucose and adherence). In order to account for affect and its impact on glucose in closed-loop systems, future studies using CGM technology can provide insight into this relationship as well as shed light on the individual differences in metabolic reactivity based on glycemic levels.