The purpose of this research was to identify barriers to adolescent cell phone use, determine the feasibility of using cell phones to measure adherence in type 1 diabetes, describe how adolescents use the cell phone for diabetes, relate ecological momentary assessment (EMA) to traditional self-report and glycemic control, and identify patterns of adherence by time of day and over time. The study is novel because it is the first to measure pediatric diabetes adherence using EMA.
Overall, adolescents reported limited use of their phones to support diabetes self-care. Simple functions, such as setting reminders, were infrequent, but may also be limited by cell phone functionality, and/or school cell phone policies. There were few times when adolescents were not allowed to use their phones at home. However, one-third of adolescents were not allowed to use their phones for diabetes at school, and another one-third was not sure about school policies. School cell phone policies most likely did not impact this study for several reasons: adolescents were asked to schedule call times when they were available to answer the phone; the afternoon calling period did not have a greater portion of missed calls; and the majority of afternoon calls were scheduled after school. This restriction at school likely resulted in reporting afternoon adherence behaviors several hours after they occurred. While the timing of the present EMA study allowed for a delay between behavior and reporting of it, school policies may impact the use of cell phones to measure self-care behaviors. Given that school-based barriers to adherence are so relevant for adolescents (Nabors, Troillett, Nash, & Masiulis, 2005
), it will be useful to work with school systems and professionals to adapt policies for students with chronic illnesses.
The duration and timing of the calls used in this study appear to provide a useful template for assessment of type 1 adherence. Data were obtained for just over half of the EMA assessments, and response rates remained stable over the 10-day calling period. Technical issues resulted in approximately 19% data loss defined as call records with incomplete data. The rate of call records with partially missing data included dropped calls or hang ups. The rates of those specific events could not be quantified. Comparison of this level of incomplete data to previous research is difficult. Many previous studies have successfully utilized such calling systems for health assessment in adult and adolescent populations (Estabrooks & Smith-Ray, 2008
; Piette, 2000
; Reidel, Tamblyn, Patel, & Huang, 2008
), but fewer have specifically addressed adherence or used the systems with cell phones (Reidel, et al., 2008
; Winland-Brown & Valiante, 2000
). Allowing adolescents to call back in to the system resulted in a greater number of assessments. Adherence rates, as reported through outgoing calls, were equal to adherence reported through incoming calls. Although on average, there were an adequate number of observations per participant, the number varied considerably across participants. Some analyses were limited to those adolescents with at least 4 assessments over different days. However, it is unclear how many assessments are needed to provide maximal sensitivity and adequate sampling of type 1 diabetes adherence behaviors.
Mornings were the most challenging time of day for blood glucose monitoring and insulin administration. This insight regarding a vulnerable time of day provides a critical leverage point for the clinical improvement glycemic control in adolescents. Evening times were associated with the highest levels of adherence. Improved adherence may be related to predictable family and home routines in this population (Greening, et al., 2007
). Momentary and mobile interventions have great potential to be tailored to individually relevant times of day, and provide situational prompts and skills needed to address barriers to adherence (e.g., time management and planning).
Trajectory analyses replicated previous cross-sectional research identifying an “inadequate” self-management style (Schneider, et al., 2007
). Glycemic control for both of the “variable” groups was worse compared to the “adequate” groups. The proportion of adolescents reporting adequate blood glucose monitoring frequency was higher than that for insulin administration. Adolescents appeared more willing to complete blood glucose monitoring, which did not always appropriately lead to insulin administration. The difference in glycemic control between the insulin administration trajectory groups was clinically significant. Missed insulin doses may be one of the most impactful of self-care behaviors on glycemic control (Olinder, Kernell, & Smide, 2009
). Each trajectory group had similar EMA response rates, so the trajectory group representing “variable” adherence does not appear to be biased by more missing (and possibly less adherent) data. The patterns of adherence found here indicate that diabetes researchers and clinicians could more efficiently and effectively direct intervention resources by identifying patterns of adherence in subgroups over time.
EMA-reported adherence was not correlated with glycemic control. Trajectory analyses did indicate worse glycemic control for the “variable” adherence groups. It is important to note that the number of calls per participant with data was related to glycemic control. This finding confirms related research that has found the number of values in blood glucose meters is a valuable indicator of adherence and is related to glycemic control (Guilfoyle, et al., In press
). The EMA method paradoxically provides greater opportunity to opt out of measurement and, thus, potentially greater bias toward reporting positive adherence behaviors (Shiffman, et al., 2008
). There are several indications that did not occur in this study: there was no difference in reported adherence through inbound versus outbound calls; there were equal numbers of calls per participant between the trajectory subgroups; and, although missing call data was greater for the morning period, reports of missed blood checks and insulin doses were also worse for that time period, as opposed to being higher, and representing a response bias.
The EMA-reported adherence was moderately related to a traditional questionnaire report. Comparison between traditional self-report measures and EMA-reported adherence is complicated by the fact that established measures are typically multi-dimensional, and assess a broad range of behaviors. Items were selected from a standardized adherence questionnaire to more closely match the behaviors assessed in the EMA portion of the study. However, time frames used for responses on questionnaires, wording of items, and metrics for the response formats often do not lend themselves to direct comparison of specific behaviors. Future research that incorporates an EMA system for adherence will need to weigh the potential benefits of this method against traditional self-report. Participants needed ubiquitous access to a cell phone and a reliable connection to avoid dropped calls. At times, circumstances may have caused delays in response, lack of response, or hang ups. While the 10-day calling period tested here showed no decline in response rates, patient engagement over longer periods of time with EMA systems may be limited (Hanauer, et al., 2009
This research has several limitations. The smaller sample for the EMA limited the size of trajectory groups, and although type 1 diabetes is primarily a Caucasian disease, there was a somewhat higher rate of Caucasian adolescents in this sample than in previous studies (Hood, Butler, Anderson, & Laffel, 2007
; Nansel et al., 2009
; Schneider, et al., 2007
). Response rates might have been higher for simple text-based assessments. Text messaging was not used here because of limited flexibility in assessing multiple behaviors through one communication. Additional insights could have been gained by coding day of the week, coding intentional versus unintentional missed adherence tasks, and including carbohydrate counting in the assessment calls. At times, call scheduling required a time lag between the actual adherence behavior and report of it. That time lag was unavoidable, given the need to use outbound calls to collect data during relevant time intervals, but that lag was not quantified in this study. Finally, an assumption incorporated into the EMA assessment was that adolescents knew that they should check their blood glucose several times per day around meals.
Ecological momentary assessment provides unique opportunities for improved measurement and understanding of adherence. Results support the feasibility and utility of this method. New insights were gained into patterns of adolescent adherence that provide leverage points for targeting or tailoring interventions to improve self-care. Results should be confirmed in a larger sample and expanded to identify situational variables that influence adherence, and additional research conducted to determine the feasibility of implementing EMA in clinical practice.