Sample and Setting
The study participants were recruited from the camp counselors (N=38) who, along with the camp health care providers, would be interacting closely with the campers during their 2-week camp stay. The Barton Center for Diabetes Education runs two camp sites: one for boys (formerly known as the Joslin Diabetes Camp) and one for girls (Barton Camp). Both are located in central Massachusetts and share a coordinated health care team and supervisory staff. Orientation and training for both sites took place at the girls’ camp. Because this is a pilot study, with a focus on obtaining preliminary data on feasibility and efficacy, we did not undertake a power analysis. Of those counselors approached, 36 agreed to participate. We conducted the study first thing in the morning when the volunteers were the most alert and interactive.
Since episodes of hypoglycemia are of big concern to counselors, parents and the children themselves, we decided to focus on its management and treatment. The hypoglycemia vignette included a two-part description for both mild-moderate and moderate-severe episodes of hypoglycemia based on comparable activities that could occur at camp. The vignette described a school-aged child playing soccer at camp who complained of feeling dizzy. The vignette prompted participants to think about what should be done first, why, what is going on, and how to prevent the situation from occurring in the future. For each of the hypoglycemia descriptions there were objectives for the diabetes educator who was leading the teaching session to follow. For instance, for the mild-moderate episode of hypoglycemia we included objectives such as the staff should be able to state signs and symptoms of hypoglycemia (mild to moderate); causes (precipitating factors) of hypoglycemia; three types of treatment; when blood glucose level should be rechecked and when and if treatment should be repeated; and when a snack of complex carbohydrate should be used in the presence of mild to moderate hypoglycemia. We also simulated hypoglycemic tremors and how to treat the critical incident.
The diabetes educator in the experimental group used the simulator to explain, review and illustrate the content described in the objectives. The simulator has a voice-over response (I don’t feel well, my stomach hurts), a bleeding finger for practicing glucose monitoring, fat pads for practicing insulin injections, and could simulate tremors/seizures for caregiver response practice. All these capabilities were threaded throughout the vignette which lasted approximately 1 1/2 hours. The same vignette and objectives were used by another Diabetes Educator with the control group, minus the visualization with the simulator, and lasting an equivalent teaching time duration.
Data were collected on demographic characteristics (race, educational level, ethnicity, whether or not they had T1DM or had a sibling or close relative that they had cared for with T1DM). At baseline, data were collected on their diabetes knowledge and self-efficacy. Diabetes Awareness and Reasoning Test-Parents (DART-P) (Heidgerken et al., 2007
) is a 47-item multiple-choice (4 choices) questionnaire that measures diabetes knowledge and was developed for children and parents. We removed items that were not pertinent to the hypoglycemia teaching vignette for this study (such as school-related management, specific types of insulin) reducing it to 25 items. The maximum score was 25 indicating correct answers on all items. The reported Cronbach’s alpha by the developers was .92 and for this study and with the modified version it was .81.
The Self-Efficacy for Diabetes (SED) is a 22-item (Likert scale, 1=very sure I can’t , to 5= very sure I can, with higher scores=more confidence) instrument that originally measured parents’ confidence in managing adolescents care (Grossman, Brinks, & Hauser, 1987
) and was adapted to measure parents’ confidence in caring for children with T1DM (Streisand, Swift, Wickmark, Chen, & Holmes, 2005
). In our study we only used 12 (thus, a maximum score of 60) of the 22 adapted items that were pertinent to the camp caregiver perspective to measure confidence in specific tasks and skills associated with diabetes care, i.e. how confident the staff was that he or she could perform day-to-day diabetes management (eliminating items such as parents’ confidence in discussing care with health care providers). The adapted 22-item Likert scale had a reported Cronbach’s alpha of .87 and for this study with 12 items it was .94.
After IRB approval was secured from The University of Massachusetts, Worcester and at the camp, participants were recruited on the first day of orientation by the camp director. They were all given the option to participate and be selected for one of two groups. All but two agreed to participate in the study. The two males who did not want to participate were taught with the standard control education session group but did not complete the questionnaires. They chose not to participate because they didn’t want to fill out the forms. After each participant completed the informed consents and baseline questionnaires, the camp medical director had them line up and sequentially assign them to alternate between group A (the control group, n=15) and group B (the experimental group, n=21). The 1 1/2 hour teaching sessions were conducted in two separate locations in the camp. Post-education session the participants completed the same questionnaires.
Data management and analysis
Data were double entered into SPSS 14.0 then converted into a SAS dataset and analyzed using SAS version 9.2. (SAS Institute, 2008
)(SAS Institute, 2008
) Continuous variables were summarized by means and standard deviations, and discrete variables by frequencies and percents. The two treatment groups were compared regarding baseline characteristics using chi-square tests for categorical variables and Wilcoxon two-sample rank testing for continuous variables. The primary analytic approach was analysis of covariance (ANCOVA) for within-person change (post minus pre) for each outcome (total DART score and total SED score) separately as a function of treatment arm, adjusting for pre score in order to account for possible regression to the mean (Chuang-Stein & Tong, 1997
). Adjustment for baseline characteristics related to change in outcomes (Pocock, 2002
) had little impact on treatment-related differences; thus, results adjusted only for the baseline outcome value are presented.
In addition to these analyses, we conducted two sets of analyses to assess the possible impact of ceiling effects, where a respondent’s baseline response was already at the maximum value for the outcome scale and thus improvement was not possible. In particular, this was a concern for the DART in participants with previous diabetes exposure. First, we used Tobit modeling (Greene, 1993
) with right- rather than leftcensoring. In other words, a DART score of 25 and a SED score of 60 were treated as censored, i.e., all we know is that the “true” score on a hypothetical scale that would allow for larger scores is at least as large as 25 or 60, respectively. Second, the two treatment arms were compared regarding mean within-participant change using a method applied by Evans et al. which accounts for the fact that the possible range of change from pre- to post-intervention depends on the participant’s pre-intervention score (Evans, Beckett, Albert, & al., 1993
). Briefly, within-participant changes in the outcome were ranked separately for each pre-intervention score, the ranks were transformed using normal scores in order to make change scores comparable regardless of the pre-intervention score, and the two treatment arms were compared regarding these transformed ranks using Wilcoxon signed rank testing.
Finally, exploratory analyses were conducted to test for interaction (effect modification) between treatment arm and previous diabetes exposure, to test whether treatment-related difference in change was larger in those with no previous exposure.
displays the demographic data for the 36 participants, with 10 males and 24 females (2 chose not to denote gender). The majority of participants identified as white, one black, one Asian and two reported ‘other.’ In both the study arms, the majority (n=25) had some previous diabetes exposure. Mean age was 19 years in the control group, and 20 years in the experimental group. With the exception of education – a higher percentage of the experimental group had ever attended college –there were no significant differences between groups on demographics. Nor did the groups differ regarding baseline DART or SED scores (p=.9360 and .5939 respectively).
Baseline characteristics of participants by treatment arm
For aim 1 we were able to demonstrate the ease of teaching a diabetes-related vignette to camp caregivers, making it very action-oriented. According to the camp director, it generated more discussion than the usual passive teaching strategy, i.e. teacher talking with students listening. The camp staff were intrigued by the patient simulator and enjoyed participating in the educational session.
For aim 2, displays between pre and post group means as well as baseline-adjusted within-participant change in DART and SED by treatment arm. Camp participants in the experimental arm had a significantly larger change in DART scores than did control participants. Within-participant change in SED scores however, did not differ significantly by treatment arm. For both outcomes, conclusions are similar for ANCOVA, Tobit, and the ranks approach (results are not shown for Tobit and ranks for the sake of brevity).
Within-Participant Change in DART and SED, Adjusted for Baseline Outcome Value, by Treatment Arm
presents results of exploratory ANCOVA for aim 3, testing the hypothesis that experimental-related differences in DART and SED change were larger in the subgroup with no previous diabetes exposure. For both outcomes, experimental participants’ mean change was significantly larger than control participants’ mean change in the subgroup with no previous diabetes exposure, but smaller and not statistically significant in the subgroup with any previous diabetes exposure, as hypothesized. Moreover, the interaction between treatment arm and previous diabetes exposure was statistically significant for both outcomes, suggesting the presence of effect modification by previous diabetes exposure.
Within-Participant Change in DART and SED, Adjusted for Baseline Outcome Value, by Previous Diabetes Exposure Within Treatment Arm
Several caveats need to be noted, however. First, the number of participants with no previous diabetes exposure is very small – three in the control arm and four in the experimental arm. Second, likely due to these small sample sizes, one of the observations in the experimental arm with no previous exposure had a relatively high Cook’s distance (0.65), indicating possible undue influence on the model estimates, although its value did not reach the cutoff of 1.0 (Field, 2005
). Re-running the analyses removing this observation, however, yielded similar conclusions (data not shown). Third, in the subgroup with no previous diabetes exposure, the control participants’ scores at baseline were higher than the experimental participants’ scores for both the DART and the SED, consistent with a larger ceiling effect – and corresponding smaller within-participant change – for the control participants than for the experimental participants. Results from Tobit analyses, however, were consistent (data not shown). Fourth, the ranks approach did not find statistically significant treatment-related differences in the subgroup with no previous exposure, perhaps because of the small sample sizes as well as the use of ranks, which can attenuate the impact of large values. In sum, these results regarding effect modification must be considered exploratory in nature.