Skills Lacking in Estimating Carbohydrate Content: A Need for Continual Education of Adults with Type 1 Diabetes
Keywords: Type 1 diabetes mellitus; Glycaemic control; Carbohydrate counting; Carbohydrate estimation accuracy; Patient education
The total amount of carbohydrates consumed in a meal is the major predictor of the postprandial glucose response [1,2]. Consequently, national and international clinical guidelines recommend that people with type 1 diabetes (T1D) learn basic carbohydrate counting (BCC) or similar experience-based methods as a meal-planning approach. BCC involves understanding the relationship between food and plasma glucose levels with special attention to the type, amount, and distribution of carbohydrate-containing foods [3,4].
Monitoring the carbohydrate intake is of paramount importance in the day-to-day regulation of especially postprandial blood glucose levels and thereby also long-term glycaemic control [5-10].
Studies have shown that children and adolescents with T1D tend to assess their intake of carbohydrates inaccurately and their lack of skills in estimating the carbohydrate content have been associated with higher blood glucose variability [11] and deteriorated glycated haemoglobin A1c (HbA1c) [12-14].
Not all Danish patients with T1D treated at outpatient clinics are referred to routine visits with a dietitian for nutritional education. However, in Denmark it is mandatory that all patients with T1D receive education in BCC in a diabetes outpatient clinic before and after commencement of insulin pump therapy. This includes training skills in gram counting of carbohydrates and experience-based estimation of high-carbohydrate foods, so that patients gradually become more aware of how to estimate carbohydrate portion sizes accurately using their eyesight. We hypothesized that previously educated adults with T1D would be highly skilled in identifying carbohydrates and estimating carbohydrate content of foods.
The primary aim of the study was to investigate skills in BCC (i.e. identifying carbohydrate sources and estimating carbohydrate portion sizes) in adults with T1D on insulin pump therapy. The secondary aim was to examine whether skills in accurate estimation of carbohydrate content of foods are associated with lower HbA1c levels.
Adults with T1D from the outpatient clinic at Steno Diabetes Center Copenhagen were invited to participate in the study. All data were collected between April and June 2013. The inclusion criteria were adults (18–75 years) with T1D on insulin pump therapy and having attended at least one consultation with a dietician within the last two years. Exclusion criteria were carbohydrate counting naive, visually impaired and people diagnosed with T1D less than 12 months before beginning of the study. Informed consent to participate in the study was obtained from all participants according to Danish regulations. The study was a quality assurance project approved by the local ethics committee, the Danish Data Protection Agency which included an approval of an External Data Manager Agreement with the Capital Region for data storing.
Participants were asked to identify high- and low-carbohydrate sources by answering ‘yes’ or ‘no’ to whether 32 different food items contained an amount of carbohydrates which could affect their blood glucose. The 32 food items were shown on a computer screen as digital images and were developed and photographed according to practical guidelines from the National Food Institute at the Technical University of Denmark [15]. Subsequently, the participants were presented to a serving of real food and asked to estimate the carbohydrate content in grams in the presented serving without the use of aids. This procedure was repeated for 20 commonly consumed Danish foods high in carbohydrates. The study dietitians were responsible for portioning the correct amount of food (i.e. carbohydrates) in grams in each serving according to the curriculum for calculation of the error estimates.
Age, gender, diabetes duration, height and weight, HbA1c, number of years on insulin pump therapy and number of consultations with a dietician within the last two years were extracted from the patients’ electronic medical records.
Differences between study participants and non-respondents were tested using the Mann-Whitney U test for numerical data while categorical data were compared by Pearson’s Chi square test for differences in proportions.
The binomial data obtained from the participants’ skills in 57-062.
identifying high-carbohydrate foods were assessed as percentages of correct and wrong answers for each of the 32 different food items. To assess estimation accuracy the participants’ error estimates of carbohydrate portion sizes of the 20 servings of real food were calculated from the following equation:
Error estimate (g) = Estimated carbohydrate content (g) – Actual carbohydrate content (g) A negative difference was considered to indicate underestimation of the presented food.
All error estimates were non-normally distributed and thus presented as medians with interquartile range (IQR). A One-sample Wilcoxon signed rank test was performed to test whether the participants’ carbohydrate estimations differed significantly from the actual carbohydrate content of the 20 different food items. In addition, error estimates from participants who reported never or rarely eating the specific food were excluded in the analysis of differences between estimated and actual carbohydrate content.
The relationship between specific patient characteristics (age, gender, body mass index (BMI: kg/m ), HbA1c) and the participants’ overall accuracy level was determined. The participants’ overall accuracy level was defined as accurate if they had a median estimation error within ±5g, as overestimated and underestimated with a median estimation error above 5g or below -5g, respectively. Differences between the three groups of participants were assessed using a Kruskal-Wallis H test for not-normally distributed data, one-way ANOVA for normally distributed data and by Pearson’s Chi square test for differences in proportions for categorical data (gender).
For all statistical tests a two-sided significance level of p <0.05 was used. All statistical analyses were performed with the SPSS software for Windows, version 22.0 (IBM Corp, Armonk, NY, USA.
Twenty-eight % (n=96) of the 348 invited patients accepted to participate in the study (Table 1). The participants had a median age of 49 years (IQR 41-59), the median BMI was 26.0 kg/m (IQR 23.0- 29.0), and the median HbA1c was 7.5%, IQR 7.0-8.2 (59.0 mmol/mol, IQR 53.0-66.0). Median diabetes duration was 22 years (IQR 13-36) and median number of years on insulin pump therapy was 2.3 years (IQR 1-4). The participants had attended a median number of three consultations with a dietician within the last two years. The only difference between study participants and non-responders was an age difference of approximately 10 years (p<0.001, Table 1) in favor of the participants.
A very high percentage of correct answers was observed when the participants classified the 32 presented food items as either highor low-carbohydrate sources. Cream was the only food with a high percentage of wrong answers (79%) compared with the other food items where the average of correct answers was 90% ranging from 54-100% (Figure 1).
As demonstrated in Table 2 the participants were only able to perform a correct estimation of the carbohydrate content in three of the 20 food items investigated (i.e. raisins, pizza, and boiled rice, Table 2). In 10 of the 20 food items the participants significantly overestimated the carbohydrate content, and in 7 of 20 they significantly underestimated estimated the carbohydrate content (p<0.001; Table 2).
Only age differed significantly between the three groups (accurate, BMI: body mass index; F: Female; M: Male; IU: International Unit; y: years. Data are medians, IQR: 25 0.05 significant using either Pearson’s Chi square test for differences in proportions (a) or Mann-Whitney U test (b).
percentile;P< 0.05 significant using One-sample Wilcoxon signed rank test.
underestimation, or overestimation) categorised by their median accuracy level when estimating the 20 high-carbohydrate food items (p=0.014; Table 3).
To our knowledge our study is the first to examine how skilled adults with T1D are in identifying and estimating high-carbohydrate foods (i.e. the concept of BCC). We found that adults with T1D were highly skilled in identifying commonly eaten high-carbohydrate foods but lacked skills in estimating the carbohydrate content correctly in 17 of 20 food servings of high-carbohydrate foods.
In addition to the high percentage of correct answers in
Gender (M/F), % (n) 96 45.8/54.2 (44/52) Age, y 96 49.0, 41.0-58.8 BMI, kg/m 96 26.0, 23.0-29.0 (59.0, 53.0-66.0) Years with diabetes, y 96 21.5, 13.0-35.8 Total insulin dosage, IU 96 36.0, 27.0-49.5
= Overestimation = Accurate estimation = Underestimation th th Data are medians, IQR: 25 -75 57-062.
identifying almost all high-carbohydrate food items, high percentages of wrong answers were only observed for cream and to a minor extent for walnuts, avocado, and pork crackling. These foods have relatively low carbohydrate content and a high content of dietary fat and thus the high percentages of wrong answers may reflect difficulties in distinguishing between foods’ calorie content and carbohydrate content. A finding consistent with a study [16] where children with T1D accompanied by their parents, and adolescents with T1D also had a tendency to estimate the carbohydrate content inaccurately in foods with a low-carbohydrate/high-calorie content (and thus a high fat content). Another explanation for high percentages of wrong answers for cream, walnuts, avocado, and pork crackling maybe that
a 252 32.9/67.1 (83/169) 0.026 b 252 39.5, 30.0-48.0 <0.001 b 250 25.0, 23.0-29.0 0.318 b 252 7.5, 6.8-8.3 0.804 (59.0, 51.2-66.8) b 247 23.0, 15.0-34.0 0.386 b 252 2.4, 1.0-4.0 0.941 b 252 38.0, 30.0-49.0 0.298 b 252 2.0, 0.0-4.0 0.342 th th -75 percentile or proportion (numbers); P <
96 5.0, 5.0-10.0 <0.001 96 6.0, 1.0-16.0 <0.001 96 5.0, 0.0-10.0 <0.001 96 10.0, 5.0-20.0 <0.001 96 5.0, -5.0-15.0 <0.001 96 10.0, 0.0-20.0 <0.001 96 6.5, 0.0-15.0 <0.001 96 5.0, 0.0-20.0 <0.001 96 10.0, 0.0-20.0 <0.001 96 10.0, 10.0-20.0 <0.001 96 -1.0, -5.0-5.0 0.118 96 0.0, -10.0-10.0 0.744 96 0.0, -10.0-20.0 0.475 96 -5.0, -10.0-5.0 <0.001 96 -10.0, -20.0-5.0 <0.001 96 -10.0, -10.0-(-5.0) <0.001 96 -10.0, -20.0-7.5 <0.001 96 -5.0, -10.0-0.0 <0.001 96 -10.0, -15.0-3.8 0.007 96 -14.0, -28.8-(-5.0) <0.001 significant using either Pearson’s Chi square test for differences in proportions (a), one-way ANOVA (b) or Kruskal-Wallis H test (c).
No of dietician visits within the last 87 3.0, 0.0-4.3 2 y F:Female; M: Male; IU: International Unit; y: years. Data are medians, IQR: 25 these foods are not that frequently eaten compared with lasagne, banana, rye bread, pasta, biscuit, and potatoes – food items that all participants were able to identify as high-carbohydrate containing food items. Interestingly, lasagne was also the food item that our participants underestimated the most with a median error of 14g in estimated vs. actual carbohydrate content. The reason is probably the relative complexity of lasagne, which in many ways can be considered as a mixed meal since it consists of several ingredients. This might have influenced the participants’ ability to estimate the high-carbohydrate ingredient (i.e. the pasta sheets) resulting in an underestimation of the carbohydrate content. Challenges with mixed meals complicating the ability to estimate carbohydrates is consistent with a study [12] in which adolescents with T1D significantly under- or overestimated the carbohydrate content when foods were presented as mixed meals.
We also observed contrasting results for boiled and sliced potatoes which most of our participants mainly over- and underestimated, respectively. Although potatoes intuitively appear easy to assess (i.e. the number and size of the potatoes) the irregular shape of the not-separated potato slices may influence the perception of the portion size [17,18] and hence complicate the assessment of the total carbohydrate content presented [19]. Another pitfall in relation to potatoes is that the carbohydrate content increases approximately by one third when potatoes are being sliced and prepared in the oven as compared to boiled potatoes [20]. The difficulties in our study in estimating portion sizes of potatoes is, however, consistent with previous studies when the potatoes were presented either as whole potatoes [17,21] or as potato slices [16].
We found that the majority of all food items in the present study were significantly over- or underestimated, which is in line with two previous studies reporting that adolescents with T1D over- or underestimated all the presented food items with 47-55% and 17-25%, respectively [12,13]. Other studies also confirm a high prevalence of carbohydrate estimation inaccuracy in adults with T1D who either underestimated 63% of all meals with more than 20% of the actual carbohydrate content [11] or 82% of all patients overestimated the carbohydrate content by 40% in average [22]. In contrast to our study and others [11-13,22] that rigorously reported the level of overand underestimation, several studies have not quantified the degree of accuracy but only concluded that the participants estimated the 57-062.
c 3.0, 1.0-4.5 3.0, 0.0-6.3 0.910 th th -75 percentile, mean ±SD or proportion (numbers); P < 0.05 carbohydrate content accurately [14,23,24] or that large variations in estimations were observed [25].
In relation to glycaemic control we did not find any significant difference in HbA1c between participants across categories of overall accuracy level. Still, the group with an accurate carbohydrate estimation (median estimation error within ±5g) had the lowest HbA1c (7.4%, 57.5 mmol/mol) compared with the group which underestimated (HbA1c 7.7%, 60.5 mmol/mol) or the group which overestimated the overall accuracy level (HbA1c 8.1%, 64.5 mmol/mol). Our result is consistent with a previous randomised control trial [12] in which no difference in HbA1c was found between adolescents who did and did not estimate carbohydrates correctly. In contrast, other studies confirm a significantly lower HbA1c in children who had good knowledge about carbohydrate counting [26] and in adolescents who estimated the carbohydrate content in dinner meals accurately [13]. However, in the latter study, HbA1c was not significantly different between adolescents when a whole day’s intake including lunch and snacks were assessed. Several other studies did not find any differences in HbA1c in adults who obtained a high carbohydrate accuracy test score [22], or attended a carbohydrate counting training program [24], or when children (accompanied by caregivers) and adolescents estimated carbohydrates accurately [27]. Clearly, lack of consensus exists among studies focusing on carbohydrate counting and its effect on HbA1c. In addition, the lack of standardisation in determining estimation accuracy is inconvenient in studies focusing on portion size assessment and it remains difficult to compare results between studies and thus prove the clinically importance of educating patients in estimating the carbohydrate content accurately for obtaining optimal glycaemic control. Thus, our cut-off for estimation accuracy might be a limitation (less than ±5g of the actual carbohydrate content) since this definition equate a participant with an estimation error of i.e. ±6g with one making estimation errors of i.e. ±50g – estimation errors with very different impact on blood excursions and hence the risk of complications. Our cut-off for estimation accuracy has been applied in previous studies [13,22], but other definitions of carbohydrate estimation accuracy have also been used [12,16]. Two studies [28,29] have found that carbohydrate estimations within 10 grams of the actual carbohydrate content in a meal are sufficient to maintain a postprandial blood glucose within the target range in children and adolescents with T1D and thus argue in favour of a cut-off for estimation accuracy of 10g instead of 5g. Nevertheless, we used a cut-off of 5g for our participants since they are advised to take additional mealtime insulin to cover carbohydrate intake as low as 5g due to the flexibility and ease of insulin matching with an insulin pump and the effect on glycaemic control with more tight insulin-to-carbohydrate matching. In addition, our participants had been trained by a skilled dietician in correct carbohydrate estimation before receiving an insulin pump as compared to other people with T1D to whom this training is still not part of the standard dietary treatment. However, at the time when the study was conducted we did not offer our patients the same intensive education and hands-on training in BCC including carbohydrate estimation as we do today, but again this intensive hands-on training is still not part of the routine nutrition education at outpatient diabetes clinics in Denmark or abroad and we are currently examining the effect on glycaemic control in a randomized controlled trial with adults with T1D on MDI therapy with a basal-bolus insulin regime [30]. Nevertheless, the preliminary results in this observational exploratory study indicate that the observed estimation errors can be explained by inadequate carbohydrate estimation skills despite some training in BCC.
Another limitation of our study is the included pre-weighted portion sizes of pre-selected foods since the selected food items, as well as the presented portion sizes, may have deviated from the participants’ usual dietary habits, thereby influencing the participants’ skills in estimating the presented portion sizes correctly. Nevertheless, we did try to take this into account by recording information about the participants’ eating frequency of the presented food items and excluding carbohydrate estimates from participants who reported never or rarely eating the specific food item.
Certainly, we have demonstrated that carbohydrate counting is a difficult task [31] which emphasizes the need for continual assessment of skills and repeatedly delivering nutrition education and support in carbohydrate counting in people with T1D to achieve and maintain skills in correct carbohydrate counting estimation [3]. See Table 1 in [32] for data on the use of a simple tool for assessment of carbohydrate estimation skills in adults with T1D [32].
We found that adults with T1D are very good at distinguishing between food items with or without carbohydrates but do not have a similar ability to accurately estimate carbohydrate portion sizes. The participants in our study who assessed carbohydrate content with the highest accuracy (within 5g) tended to have the lowest HbA1c. Our results underline the importance of continual education of people with T1D in BCC by skilled dieticians.
This work was supported by funds for research and development in the clinic at Steno Diabetes Center Copenhagen, Denmark. BE planned the study design and was responsible for carrying out the study. AG collected the data, NS conducted the statistical analyses, and wrote the first draft of the manuscript. All authors were involved in the data analysis and interpretation and revised the manuscript and approved the final version.