- Email: [email protected]
Outcomes of educational interventions in type 2 diabetes: WEKA data-mining analysis
Patient Education and Counseling 67 (2007) 21–31 www.elsevier.com/locate/pateducou
Review
Outcomes of educational interventions in type 2 diabetes: WEKA data-mining analysis Arun K. Sigurdardottir a,*, Helga Jonsdottir b, Rafn Benediktsson c,d,e a
Faculty of Health Sciences, University of Akureyri, 600 Akureyri, Iceland b Faculty of Nursing, University of Iceland, Iceland c Department of Endocrinology and Metabolism, Landspitali University Hospital, Iceland d Faculty of Medicine, University of Iceland, Iceland e Icelandic Heart Association Iceland, Iceland Received 30 November 2006; received in revised form 4 March 2007; accepted 6 March 2007
Abstract Objective: To analyze which factors contribute to improvement in glycemic control in educational interventions in type 2 diabetes reported in randomized controlled trials (RCT) published in 2001–2005. Methods: Papers were extracted from Medline and Scopus using educational intervention and adults with type 2 diabetes as keywords. Inclusion criteria were RCT design. Data were analyzed with a data-mining program. Results: Of 464 titles extracted, 21 articles reporting 18 studies met the inclusion criteria. Data mining showed that for initial glycosylated hemoglobin (HbA1c) level 7.9% the diabetes education intervention achieved a small change in HbA1c level, or from +0.1 to 0.7%. For initial HbA1c 8.0%, a significant drop in HbA1c level of 0.8–2.5% was found. Data mining indicated that duration, educational content and intensity of education did not predict changes in HbA1c levels. Conclusion: Initial HbA1c level is the single most important factor affecting improvements in glycemic control in response to patient education. Data mining is an appropriate and sufficiently sensitive method to analyze outcomes of educational interventions. Diversity in conceptualization of interventions and diversity of instruments used for outcome measurements could have hampered actual discovery of effective educational practices. Practice implications: Participation in educational interventions generally seems to benefit people with type 2 diabetes. Use of standardized instruments is encouraged as it gives better opportunities to identify conclusive results with consequent development of clinical guidelines. # 2007 Elsevier Ireland Ltd. All rights reserved. Keywords: Educational interventions; Type 2 diabetes; Randomized controlled trials
1. Introduction Prevalence of type 2 diabetes is increasing [1]. It is a costly condition which causes significant morbidity and mortality. Good metabolic control clearly reduces the occurrence of complications [2]. Metabolic control rests heavily on self-care of the individuals having diabetes as well as their families. Since knowledge is a prerequisite for effective self-care, diabetes education has become an integral part of diabetes care [3]. The goal of the education is to enhance diabetes related self-care that contributes to good metabolic control and well-
* Corresponding author. Tel.: +354 460 8464; fax: +354 460 9999. E-mail address: [email protected] (A.K. Sigurdardottir). 0738-3991/$ – see front matter # 2007 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.pec.2007.03.007
being, which hopefully minimizes the occurrence of both acute and chronic complications. Several studies have been carried out on diabetes educational interventions. However, standardization of outcome measures is in its infancy. In a systematic review, Norris et al. [4] reported that ‘‘knowledge was measured via a variety of instruments often lacking in reported reliability and validity’’ and it seems that incorrect information or knowledge regarding diabetes treatment is common among people with diabetes [5]. Systematic reviews have indicated that diabetes education, at least in the short term, increases knowledge among people with type 2 diabetes [3,4]. Deakin et al. [3] documented in a metaanalysis that group based diabetes educational programs improved diabetes knowledge for up to 14 months. Educational interventions seem to enhance positive changes in self-reported
22
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
diet such as carbohydrate and fat intake [4]. The effectiveness of education with regard to metabolic control among adults with type 2 diabetes has been confirmed in at least two systematic reviews [6,7], where the impact of education has lowered glycosylated hemoglobin (HbA1c) by about 0.3–1.4% for as long as 12–14 months following educational intervention [3,6]. Interventions teaching self-care skills seem more effective than interventions providing information only in improving clinical outcomes [4,8]. Specific self-care instructions have been shown both to increase self-efficacy and self-care where increased self-efficacy contributes to increased self-care [9]. Self-efficacy is task specific [10] and Brown [11] maintained that the focus in diabetes education should be on narrow dietary goals and exercise as it has larger effects on weight loss and HbA1c level than general broader aimed interventions. This postulation is supported by an analysis of 36 studies on diabetes education [12] where self-care skills interventions were more frequently associated with improvements in metabolic control than informational or educational interventions. Psychosocial factors such as attitude and distress affect self-care and can modify self-care [13,14]. Diabetes educational programs have been criticized for being centered on knowledge and physiological outcomes, whereas the patient’s experience of living with the disease has not been given due concern [13]. Similarly, in Norris et al. [4] meta-analysis only 5 studies out of 72 included psychosocial variables as outcome variables. Although evidence exists regarding the ability of educational interventions to improve both knowledge and metabolic control, it is still unclear what contributes directly to this improvement. By knowing which factors are most important, health care practitioners would be better able to focus their work, improve efficiency and keep costs associated with the educational programs at a minimum. The purpose of this systematic review was to analyze educational interventions conducted in published randomized control trials (RCT) on people with type 2 diabetes. We have systematically looked for conceptual models on which interventions were based on, the delivery and teaching methods, settings, content and intensity of interventions and characteristics of providers. Measurements of outcomes were also considered. Key factors influencing improvements in glycemic control were identified by using data-mining methods. 2. Methods 2.1. Databases searched for publications The searched databases were Medline, Scopus and the Cumulative Index to Nursing and Allied Health Literature (CINAHL), including year 2001 to October week 3, 2005. Previously, Norris et al. [4,6] and Ellis et al. [15] analyzed RCT studies published up to the year 2000. A librarian assisted with the database search. Terms searched were diabetes mellitus.mp or exp *Diabetes Mellitus, patient education/or educational intervention.mp, including all subheadings. The terms were also combined and limited to the English language and adults with type 2 diabetes.
Exclusion criteria were set as; diabetes type 1 or if type of diabetes was not reported in the study or if results were not analyzed according to type of diabetes. Studies regarding gestational diabetes, pilot studies and preliminary reports were excluded. Titles of articles were independently analyzed by two persons in order to ensure that relevant studies were retrieved. Reference lists of retrieved articles were also scrutinized. 2.2. Data mining Data mining and knowledge discovery in databases (KDD) was used to analyze the data. The KDD process maps low-level data that are often too voluminous to be understood into forms that are more compact or useful in business [16] or even health care. The KDD process uses specific data-mining methods (computer algorithms) for pattern discovery and extraction. Data mining involves finding patterns or rules which represent new knowledge or ideas which then can be used to make predictions or form the basis of hypotheses for future experiments [16–18]. Computer algorithms are used to find rules or decision trees that best explain a data set. A branch in a rule or decision tree occurs for particular values of an attribute (variable) and the consequent rule or the leaf nodes of a decision tree, makes the final classification for the attribute chosen as the attribute being predicted or the dependent attribute. Computer algorithms used to find rules or decision trees work by recursively selecting the attribute which does the best job of classifying (discriminating) the attribute being predicted [16,18]. This tactic is employed because it usually results in rules or decision trees which are quite small and which can be used to interpret the data set. Typically few of the attributes in the data set are selected to form the rule or decision tree [16]. A data set to be mined is described by attributes (variables) and instances in this case a single study. Numeric and nominal attributes are used to describe a data set to be mined. Nominal attributes take on values in a pre-specified, finite set of possibilities [16]. In our data set, we converted all numeric attributes into nominal attributes, see Section 2.5. Typically, one of the attributes is taken as the dependent attribute representing the concept to be predicted by a pattern or a rule [16,17]. However, it is possible to consider each attribute as the dependent attribute. Coding attributes and checking them for incorrect attribute values and typographic errors is important. Results of initial data-mining attempts can suggest that attributes should be removed or combined; such actions require that the data set is carefully prepared before the subsequent KDD process [16–18]. An instance is characterized by values of the attributes. In the data set analyzed in this paper, an instance characterizes a single study. When using data mining to predict models or rules it is important that instances are correctly predicted, or the accuracy of the prediction is high. Accuracy is the fraction of examples predicted correctly by data mining according to the model and measures how error-free the model’s predictions are [17]. Error in the model prediction is often referred to as a misclassification [16,17].
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
2.3. Data extraction When extracting factors to use in data mining, Elasy et al.’s [19] taxonomy regarding educational interventions was applied for the intervention group, with minor additions, such as whether a theory was guiding the intervention and mean age of study participants. Intensity of the intervention was calculated as the number of episodes multiplied by duration of episodes to obtain how many hours the intervention lasted. Duration of intervention in months was a separate variable in data mining. When multiple post-intervention values (glycosylated hemoglobin (HbA1c) level, body mass index (BMI)) were available the newest value was used. Difference in HbA1c level pre- and post-intervention (Idiff) is the main outcome analyzed by data mining. However, other variables were also tested as dependent attributes, including theory, provider (nurse, physician and research team), intensity of the intervention, duration of the intervention, follow-up time, follow-up time after the intervention finished, mean age and reduction in BMI. 2.4. The machine learning program WEKA Waikato Environment for Knowledge Analysis or WEKA, Version 3.4.3 was used for data mining. WEKA was developed at the University of Waikato in New Zealand [16]. The WEKA program is written in Java, is available on the Internet [20], and comprises a variety of data-mining algorithms. In this study, we made use of the J48 classifier algorithm which is an implementation of the C4.5 decision tree learner, see Witten and Frank p. 198 [16]. 2.5. Data analysis Initial data-mining attempts involved considering which attributes to code and how they should be coded by studying histograms of values and doing many data-mining attempts. Histograms of each attribute were examined to determine natural groupings of values into categories as well as finding mean values of attributes. Cutoff points occurred at natural dips in the histograms. Since data-mining attempts with three or four categories returned large rules with little meaning, the final attempt simply used two possible categories for each attribute. Using two possible categories for every attribute in this way also eliminated bias in the data-mining process towards attributes having many possible categories over attributes with few possible values. The best definitions were set as follows: The HbA1c level was coded as: The attributes Idiff and Cdiff represented the difference in HbA1c level between pre- and post-intervention for the intervention and control groups, respectively. Idiff had two possible values defined as: (a) small change, +0.1 to 0.7% (9 studies), (b) big reduction, from 0.8 to 2.5% (9 studies). Cdiff also had two possible values defined as: (a) small change from +1.2 to 0.4% (12 studies) and (b) big reduction, from 0.5 to 1.7% (5 studies).
23
Initial HbA1c level was coded as: high, 8.0% (8 studies) and low 7.9% (10 studies). Intensity of intervention was coded as: (a) less, <11 h (6 studies) and (b) more, 11 h (12 studies). Duration of intervention was defined as: (a) short, <7 months (10 studies) and (b) long, 7 months (7 studies). The follow-up assessment time was defined as: (a) less, <12 months (7 studies) and (b) more, 12 months (11 studies). Attributes representing setting, delivery, teaching method, provider and content were given yes (y) or no (n) definitions, e.g. if diet was addressed in intervention then it was classified with y. A t-test was also used to test for difference in HbA1c level between groups, to supplement the data-mining results. 3. Results The search of databases resulted in 464 publications. From Medline, 19 articles were retrieved reporting 16 studies. From Scopus, one study not reported in Medline and one study from reference lists also not reported in Medline met our inclusion criteria. As CINAHL was the last database to be searched, all articles (n = 17) reported in it had already been retrieved from the other databases. The total numbers of included studies were 18, reported in 21 articles. Fig. 1 shows the systematic review flow diagram and reasons for exclusion. 3.1. Validity assessment Internal validity was assessed by van Tulder’s et al. [21] scheme for assessment of internal validity, where high quality of a research is higher or equal to 6 on a scale from 0 to 11 [21]. Two reviewers independently assessed the internal validity of the interventions. The assessments were compared and agreed upon. In the validity assessment, 7 studies of 18 received scoring of 6 or higher, or good scoring and two studies received scoring of one, see Table 1. One study, Acik et al. [22], used three groups. Of those we excluded one group in the analysis (WEKA), the diet-exercise group, as selection into it was based on extremely high pre HbA1c level. 3.2. Educational interventions Fig. 2 shows the taxonomy of the interventions, e.g. how often theory is guiding the intervention or what is included in the interventions’ content. The taxonomy for each intervention is not mutually exclusive as for example, teaching methods for one intervention could include all four teaching styles. Table 2 is a summary table of the interventions and the main findings displayed in a chronological order. 3.3. Theory guided interventions In ten cases of 18 a theoretical framework guiding the intervention was reported. Within the theoretical frameworks
24
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
Fig. 1. Systematic review flow diagram of retrieved studies, n, number of articles.
there existed diversity. For example, Ho¨rnsten et al. [23] evaluated whether person centered intervention was more effective than conventional diabetes care. An empowerment theory guided Pibernik-Okanovi et al.’s [24] and Rickheim et al.’s [25] studies. Culturally appropriate educational
programs in minority populations in USA were the focus of four studies concerning Native Americans [26], Mexican Americans [27,28] and African–Americans [29]. When testing whether a theoretical framework affected how the intervention was organized (content), how long the
Table 1 Assessment of methodological quality; highest possible score is 11 and the lowest is 0 Reference
Acik et al. [22] 1
AndersonLoftin [29] 7
Brown et al. [28] 6
Brown et al. [27] 6
Cooper et al. [33] 4
Gaede et al. [35] 5
Reference
Miller et al. [36]
Rickheim et al. [25]
Sarkadi and Rosen-qvist [38]
Schwedes et al. [37]
Sone et al. [32]
Total quality score
4
Pibernik Okanovic et al. [24] 1
4
4
5
6
Total quality score
Ho¨rnsten et al. [23] 4
Kim et al. [31] 3
Surwit et al. [26]
Trento et al. [30]
Wolf et al. [39]
2
8
9
Goudswaard et al. [34] 8
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
25
Fig. 2. Percentage of interventions (n = 18) using each attribute in the intervention. The y-axis represents % of studies including exact attribute, where 18 = 100%. The x-axis shows attributes such as setting, provider, teaching and delivery methods and content of the interventions.
follow-up time was or difference in BMI and HbA1c level preto post-intervention no influence was found in data mining. No statistically significant difference in reduction in HbA1c level was found between theory guided interventions and non-theory guided interventions (t(16) = 0.66, p = 0.516). 3.4. Setting, delivery, provider and teaching method Group education in combination with individual education was commonly used as 13 of 18 interventions applied that form. Two studies compared group and individual education [25,30]. The most common delivery method was face to face (17 studies) and one study [31] used a combination of telecommunication and written instructions. Sone et al. [32] used three types of delivery, face to face, telecommunication and written instructions. Providers of interventions were nurses [23,33] physicians [22,30], team of health care providers [27,28,34,35] and dietitians [36]. All interventions apart from one [22] used collaborative teaching methods such as goal setting, problem solving and cognitive reframing. Differences in setting, delivery, provider and teaching method were not related to differences in HbA1c level pre- to post-intervention. 3.5. Content of interventions Content was most often reported as teaching about basic diabetes knowledge and self-care skills such as diet and exercising, medication adherence, self-monitoring of blood glucose (SMBG) and psychosocial aspects. Three studies explicitly declared that education was based on initial assessment [24,36,37]. However, it is unclear from two papers [24,37] how initial assessment was conducted and whether interventions were tailored to the assessment.
3.5.1. Knowledge Although almost all interventions included basic diabetes knowledge, measuring knowledge as an outcome was only conducted in five studies using different instruments. Brown et al. [27,28] and Trento et al. [30] tested knowledge as an outcome with instruments that they had developed but Sarkadi and Rosenqvist [38] did not detail how they measured knowledge. All the above-mentioned studies demonstrated improvements in knowledge with education, even four years after the intervention [30]. When testing if measurement of knowledge affected reduction in HbA1c level post-intervention, data mining did not detect difference and no significant difference was found (t(16) = 0.03, p = 0.490). 3.5.2. Self-care skills Self-care skills were classified into four areas as: diet or nutrition, physical activity, medication adherence and SMBG [14]. Most of the interventions included self-care areas and ten interventions described all these four self-care areas in their educational content. However, self-care skills were seldom measured and the measurements that were carried out included diverse instruments. Cooper et al. [33] measured self-care by the Diabetes Self-Care questionnaire and Anderson-Loftin et al. [29] measured dietary fat behavior by the FHQ-25 questionnaire. SMBG measurements were included in nine interventions, see Table 2. Comparison of groups that included SMBG in the intervention to those that did not demonstrated no difference in HbA1c reduction according to data mining and ttest pre- to post-intervention (t(9) = 1.72, p = 0.119). 3.5.3. Emotional aspects Of 13 studies which included teaching emotional aspects, only eight interventions measured the effect of this inclusion.
26
Table 2 Characteristics and outcomes of educational interventions included in the review Focus of paper, participants
Intervention
Outcomes
Conclusion
Gaede et al., Danmark [35]
To assess the effect of intensified education, on diet, exercise, and smoking (I: 80, C: 80)
All patients in both groups informed of basic diabetes guidelines, I: individual diet interventions and patients set their own goals. Group intervention regarding diet, exercise and smoking, C: nr (I I: more, D I: more, F: more)
Metabolic control + I, metabolic control C, dietary behaviour+, weight reduction I and C, no difference between groups in time used to exercise
Dietary behavior analyzed other self-care skills not measured. High resources and only modest changes in behavior
Miller et al., USA [36]
To evaluate nutrition intervention on blood glucose and lipoprotein levels (I: 47, C: 45)
I: 10-w sessions in healthy eating. Each session lasted 90 min. Theory of Meaningful learning and Social Cognitive Theory, C: usual care (I I: more, D I: less, F: less)
Metabolic control + I, metabolic control, nc C, treatment goals for total cholesterol+
Nutrition education improves metabolic control among older adults
Schwedes et al., Germany and Austria [37]
To investigate the effect of meal-related SMBG on glycemic control and well-being in non-insulin treated people (I: 113, C: 110)
I: instruction in use of SMBG device and measure BG six times daily, before and 1 h after main meal and record the values in a diary for BG and well-being. Seen by a nurse every 4 w, six times and asked to reflect on self-regulation and life with diabetes, C: counseling on diet and lifestyle (I I: more, D I: less, F: less)
Metabolic control + I, metabolic control + C, weight reduction + I, weight reduction + C, satisfaction with treatment I + and C, well-being + I
Keeping a diary and recording eating habits and SMBG enabled more autonomous self-care
Brown et al., USA [27]
To compare 2 interventions designed for Mexican Americans (I: 126, C: 114)
I: 2-h weekly sessions for 3-m, diet, exercising, SMBG. 6-m of 2-h biweekly and 3-m of monthly 2-h sessions to promote behavior change. 52-h contact. Family member brought to the program., C: usual care, 1-y waiting list (I I: more, D I: more, F: more)
Metabolic control + I, metabolic control C, BMI + I, BMI C, knowledge + I and C
Culturally specific education for Mexican Americans is beneficial. High resources and only modest changes in behavior
Surwit et al., USA [26]
To assess if stress management can improve glucose control (I: 60, C: 48)
C: five 30 min sessions on exercise and diet, and videos featuring how Native Americans should incorporate traditional Native American values and foods into life, I: in addition to the same education as C, there was a stress management program (I I: more, D I: less, F: more)
Metabolic control + I, metabolic control C, stress-management + I group
Intervention supports efficacy of stress management training. Self-care skills not measured
Trento et al., Italy [30]
To compared traditional individual diabetes care with a model in which routine follow-up is managed by interactive group visits (I: 56, C: 56)
I: group education very 3-m, diet, exercise, SMBG, medication and complications. Annually diabetes screening, C: visit every 3-m to same physician, knowledge assessed and education offered, diet, SMBG, exercising (I I: more, D I: more, F: more)
Metabolic control + I, metabolic control C, BMI + I, BMI + C, quality of life + I, knowledge + I and C, self-care+
Group care is feasible in an ordinary clinic and costeffective in preventing the deterioration of metabolic control and quality of life
Sone et al., Japan [32]
To asses if long term life-style intervention can improve glycemic control and prevent complications (I: 1105, C: 1100)
I: dietary habits, physical activities and adherence to treatment. 15 min telephone counseling sessions at least once every two weeks, C: usual care (I I: more, D I: nr, F: more)
Metabolic control + I and C, BMI + I, BMI nc C
No instruments used to measure QoL, well-being or self-care skills
Rickheim et al., USA [25]
To compare effectiveness of delivering diabetes education in either a group or individual setting using consistent evidence-based curriculum. (I: 87 (group education) C: 83 (individual education))
Both groups received the same education, diet, SMBG, physical activity, foot care, problem solving emphasizing to empower and meeting the needs of an adult learner, 5–7 h of education over 6-m (I I: less, D I: less, F: less)
Metabolic control + I and C, BMI + I and C, knowledge + I and C, attitude + I and C, QoL + I and C
Group education is as effective as individual education. The optimal group size is not known
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
Study
To find if intervention impacting upon illness beliefs and leading to changes in self-care behaviors affects blood glucose control (I: 53, C: 36)
8 weekly group sessions, 2-h each, diabetes nurses, experiential learning, diet, exercise, emotions, SMBG goals, behavior change, C: nr (I I: more, D I: less, F: more)
Metabolic control I, metabolic control + C, attitude + I, self-care + I and C
Reflection was useful to clarify and interpret complexities of self-care. Supportive environment was helpful and patients valued collaborative learning
Wolf et al., USA [39]
To compare efficacy of lifestyle education to usual care in primary care (I: 73, C: 74)
I: case manager met patients individually, in groups and by phone. Education, support and goal setting. Goals tailored to patients but met dietary recommendations for people with type 2 diabetes and obesity, C: usual care (I I: less, D I: more, F: more)
Metabolic control + I, metabolic control C, weight reduction + I, weight reduction C, emotional aspects + I
Change in weight did not predict change in HbA1c level. Weight gain during the last 4 months suggests need for ongoing lifestyle coaching. Self-care skills not measured
Gouds-waard et al., Netherlands [34]
To assess the long term effects of education (I: 25, C: 29)
I: collaborative education, by 2 diabetes nurses, diet, physical exercising and losing body weight, medication adherence, SMBG, BG control at home. 6 sessions over 6-m, total contact time of 2.5 h, C: usual care, 3-m reviews, focusing on diabetic symptoms and measurement of BG (I I: less, D I: less, F: more)
Metabolic control + I, metabolic control + C, weight reduction + I
Long-term effects were disappointing. HbA1c level and weight loss occurred at the same time. No instruments used to measure QoL or self-care skills
Acik et al., Turkey [22]
To analyze effects of patient education, diet and regular exercise on blood glucose control (I: 33, C: 33)
I: diet counseling C: usual care (I I: less, D I: less, F: less)
Metabolic control + I, metabolic control C, BMI + I, BMI C
No instruments used to measure QoL, well-being or self-care skills
Pibernik-Okanovic et al., Croatia [24]
To determine the feasibility of empowerment educational program on patients’ quality of life and HbA1c level (I: 73, C: 35)
I: 6 w group sessions, goal setting, problem solving, coping with diabetes, seeking social support, and staying motivated. Discussion and practical exercises, C: Usual care (I I: less, D I: less, F: less)
Metabolic control + I, metabolic control, nr C, self-care behaviors + I, quality of life + I
Patients perceived benefits from the program, support from professionals, not being criticized and stimulated for self-care
Sarkadi and Rosenqvist, Sweden [38]
To present an educational program based on participants’ experiences and use these experiences as a basis for acquisition of practical skills needed for self-management of diabetes (I: 39, C: 38)
I: 12 m group education, once in a month, where practical aspects of diabetes were analyzed by the group, such as choice and preparation of food, SMBG tasks, and how exercising decreased BG and support to deal with emotional aspects of diabetes, C: usual care (I I: more, D I: more, F: more)
Metabolic control + I, metabolic control C, knowledge+, exercising + I
Self-reported chances in weight., Assessment of exercise not other self-care skills
Brown et al., USA [28]
To compare 2 interventions designed for Mexican Americans, Program was designed according to the culture (I: 102 extended, C: 114 compressed)
I: extended, 12-w, 2-h sessions, diet, exercising, SMBG, medication adherence; 14, support group sessions, goals, problem solving 2-h each, C: compressed 8-w, 2-h support group sessions, goals, problem solving and support sessions held at 3, 6 and 12 mI (I: more, D I: more, F: more)
Metabolic control + I and C, knowledge + I and C, no significant difference found between programs
The most effective dosage of education is unknown. Selfcare skills or emotional aspects not measured
Ho¨rnsten et al., Sweden [23]
To evaluate whether a person centered intervention focusing on patients’ personal understanding of their illness is more effective than is conventional diabetes care (I: 44, C: 60)
I: 10 group sessions, 2 h each, discussion not predetermined, emanated from patient understanding and covered experiences and practical theoretical points, C: regular care in other clinics, 1–2 y visits to nurses in other clinics (I I: more, D I: more, F: more)
Metabolic control + I, metabolic control C, BMI + I, BMI C, satisfaction + I, well-being nc
No self-care skills evaluated. Focus on wellness in patients’ experiences may lead to decisions to manage the disease
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
Cooper et al., UK [33]
27
, negative change; nc, no change; nr, not reported; BG, blood glucose; SMBG, self-monitoring blood glucose; I I, intensity of I, intervention; C, control; h; hour; y, year; w, week; m, month; +, positive change; intervention; D I, duration of intervention; F, follow-up assessment.
Intervention using individualized approach through telephone improves blood glucose control and satisfaction Metabolic control + I, metabolic control C, satisfaction + I I: booklet regarding diet, exercise, medication and SMBG and keeping diet and exercise diaries, hypo- and hyperglycemia management. Nurse phone calls weekly for 12-w, C: nr (I I: less, D I: less, F: less) To investigate the effect of nurse telephone calls on HbA1c levels and adherence to self-care recommendations (I: 15, C: 10) Kim et al., Korea [31]
Culture is an important part of life and affects dietary chooses. Self-care skills or emotional aspects not measured Metabolic control + I, metabolic control + C, BMI + I, BMI C, dietary behavior + I I: four weekly classes on diet, included in social events. Four monthly 1-h peerprofessionals discussion groups, learning about diabetes. Weekly follow-up calls by nurses, C: diabetes classes for 8-h (I I: more, D I: less, F: less) To test culturally appropriate dietary self-management intervention in African– Americans (I: 49, C. 48) Anderson-Loftin et al., USA [29]
Conclusion Outcomes Focus of paper, participants
Intervention
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
Study
Table 2 (Continued )
28
Results were improvements in well-being, anxiety and attitudes towards diabetes. However, different instruments were applied; the Medical Outcome Study Short Form (SF-36) and the adjustment to diabetes instrument (ATT19) [25,39], the 22 well-being questionnaire [24,37], the WHO-QoL instrument [24], the Stage-Trait Anxiety [26] and the Diabetes Integration questionnaire [33]. Using data mining to test whether educational content affected reduction in HbA1c level pre- to post-intervention, attributes such as setting, delivery, teaching methods and provider were removed from data mining, but the results were unchanged. 3.6. Intensity and duration of intervention Most programs, or 12 of 18, used more than 11 h of intervention. The duration of interventions was from 8 weeks [22] to 12 months [28,30,34,35,39] and the follow-up assessment from 8 weeks [22] to 51 months [30]. Only seven interventions had follow-up time for more than 6 months after the intervention finished [26,30,32–35,38]. The WEKA data-mining program did not detect association between reduction in HbA1c level and these attributes. No statistical significant difference was found between groups according to follow-up time for more than 6 months after the intervention finished compared to less than 6 months on HbA1c level (t(16) = 1.57, p = 0.136). 3.7. Biological factors (HbA1c level and BMI) Pre intervention HbA1c level predicted improvement in glucose control in response to patient education. If initial HbA1c level was high or 8% the reduction was 0.8–2.5% and if the initial HbA1c level was 7.9% the change was from +0.1 to 0.7%. Studies that started with low initial HbA1c level achieved less HbA1c reduction as a response to the education. Statistically significant difference was also found between intervention groups (Idiff) in HbA1c mean reduction according to high or low initial HbA1c level (t(11) = 3.53, p = 0.004). Three studies were misclassified, e.g. the intervention achieved higher or lower HbA1c reduction than predicted according to start level. The accuracy of the prediction is 83.4% [19]. Two of the misclassified studies [22,24] had low internal validity, with scoring of 1 out of 11, see Table 1. In the third misclassified study (Goudswaard et al. [34]) the participants were all treated with maximal dosages of oral hypoglycemic agents. Goudswaard et al.’s [34] results were promising at 6 weeks as the intervention group had improved their glucose control significantly compared to the control group, but at 18 months there was no significant difference in HbA1c level as regarded the number of patients with HbA1c <7.0% or the number of patients treated with insulin. To detect if studies with low internal validity scoring would affect data-mining results, studies with internal validity 5 were removed from data mining. This left 7 interventions to be mined but the results were unchanged.
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
Fig. 3. Percentage change in glycated hemoglobin – the HbA1c level – for intervention and control group. The y-axis represents % change in HbA1c level as a response to education. Time on x-axis is number of months from beginning of the intervention. The symbol * shows significant difference between intervention and control groups in HbA1c level.
For the interventions groups the relative HbA1c level reduction is on average 6–7% compared to the control groups’, see Fig. 3. There was also a statistically significant difference in HbA1c mean reduction, between the intervention and the control groups (t(33) = 2.82, p = 0.008), even though in at least seven interventions (Table 2) the control group received more care than standard care. Seven interventions out of 18 achieved more than 10% reduction in HbA1c level and four of these interventions achieved post-HbA1c level 7.0%. Seven studies reported difference in BMI but WEKA did not demonstrate predictors of change in BMI. 4. Discussion and conclusion 4.1. Discussion Using data mining, this analysis of the literature on RCT of educational interventions found that HbA1c level before participating in the educational intervention was the main factor affecting how large an effect the intervention had on glucose control in adults with type 2 diabetes. This is in agreement with several other studies [26,28,32]. For instance, Sone et al. [32] reported that decrease in HbA1c level occurred only in patients with HbA1c level 7.5% or higher regardless of intervention or control group. As data mining has only recently
29
been used to analyze educational interventions it was decided to accompany that method with a more traditional statistics (t-test) to supplement our findings. Results from data mining and t-test were congruent. Accuracy of data mining was sufficient with three misclassifications, where two of the misclassified studies had low internal validity. Threats to internal validity were common but no study fulfilled van Tulder’s et al. criteria for methodological quality [21]. However, removing studies with low internal validity from data mining did not change the results. It should though be considered that this analysis included only 18 studies. Our findings that delivery, teaching methods or content did not influence reduction in HbA1c level, is in accordance with results from Norris et al. [4] but conflicts Ellis et al. [15] results where face to face delivery, exercise in the content and cognitive reframing explained 44% of variance in glucose control. In this review all but one intervention [36] included face-to-face delivery and didactic teaching method was applied in only one intervention [22]. All the other interventions used goal setting, cognitive reframing and support as teaching methods. That conflicts with results from Knight et al. [7] who found didactic teaching methods to be most common. The explanation could be that our review included more recent studies than Knight et al’s study. In the year 2006, Deakin et al. [40] published results from a RCT expert program among people with type 2 diabetes. Their results were concurrent with our results as the expert program increased knowledge and psychosocial adjustment to diabetes and improved metabolic control among the intervention group (n = 150), the HbA1c level reduced by 0.6% but in the control group (n = 141) it increased by 0.1%. Our results that contact time did not affect the HbA1c level conflicts results from Norris et al. [6], in which 15 studies out of 31 measured contact time according to the HbA1c level. For every additional hour of contact time, the HbA1c level was reduced by 0.04% (CI 0.04–0.08). The results that knowledge and self-care skills were rarely measured are in accordance with results from Bodenheimer et al. [8]. They analyzed 72 studies of self-management training in diabetes and found that 46 studies measured effects of patient education on patient knowledge and performance of technical skills, 33 studies showed positive impact and significant differences between groups were observed in 13 studies. However, Bodenheimer et al. [8] did not define performance of technical skills and how it was measured. Half of the interventions analyzed in this paper were explicitly guided by theoretical frameworks. As the interventions were based on different theories it is difficult to compare them and decide upon which is the most effective theory to guide diabetes education, a result that has been reported before [7]. The desired amount or intensity of education is unknown [6,8,28,41], which is supported by the fact that no difference was found in the post-intervention HbA1c level according to type or content of intervention, intensity, duration or length of intervention. Diversity of educational interventions in this review could have hampered
30
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
actual knowledge discovery in databases (KDD) of effective educational practices, which is in agreement with conclusions of other authors [6,8,15,41]. In this review, seven interventions out of 18 achieved more than 10% relative reduction in HbA1c level and the interventions groups’ HbA1c level was on average 6–7% lower than the control groups’. The control groups frequently received better care than standard care which can underestimate effects of the interventions. This is an issue that has been reported before [7] and needs to be highlighted. The studies in this review included mostly people who had been diagnosed with diabetes for longer than one year. Improvement in metabolic control in spite of the fact that the patients were well beyond the time of becoming diagnosed with the diabetes and the potential motivation which that may hold is encouraging and of particular clinical significance. Some studies showed that HbA1c level [34] increased with longer follow-up time. This may indicate that people with diabetes might benefit from periodic reinforcement of education and support even throughout life. More research is warranted in this area and also on how time from diabetes diagnosis affects result of educational interventions. Individuals in the educational programs seemed to benefit from participation, shown by increased knowledge, self-care skills and improvements in psychological aspects. However, these aspects were measured with heterogeneous instruments which make comparison difficult. Therefore the effectiveness of these aspects is not yet fully clear. This analysis included only RCT studies. Mu¨hlhauser and Berger [42] caution that systematic reviews using RCTs might exclude some studies that are beneficial, although systematic reviews are in general valuable. It is known that RCT study design is not always feasible when researching factors affecting self-care behavior among people with diabetes. Interventions intended to change behavior should focus on increased knowledge and ability of individuals to utilize the knowledge in their self-care and they need to know that psychosocial aspects can influence self-care. Such interventions have to be patient centered and tailored to the patient’s needs. Tailored patient-centered interventions can only partly be standardized [43] which makes utilization of RCT more complex and even requires other methodological approaches. 4.2. Conclusion Data mining is an appropriate and sensitive method for analyzing educational interventions. These results revealed that there exists incongruity in designing and studying educational interventions which results in difficulties when considering the best intervention to use for motivating self-care behavior among people with type 2 diabetes. Usual care was often inadequately defined and sometimes the control group was offered education that could mask the differences between the groups. The instruments used to measure outcomes were various and highlight the need for universal use of standardized validated instruments.
4.3. Practice implications Participation in educational interventions is generally beneficial for people with type 2 diabetes. People with type 2 diabetes might need support and education throughout life as effectiveness of interventions seems to lessen as time passes. People with type 2 diabetes might benefit from frequent health provider contact but the intensity does not necessarily need to be so high. We encourage development of patient centered interventions and development and use of standardized validated instruments which might facilitate identification of the best method for diabetes education and consequently the introduction of universal clinical guidelines. Acknowledgements Assistance from Dr. Andrew Brooks MACM MIEEE MBCS, Associate Professor University of Akureyri Iceland, with data mining is acknowledged. We like to thank Connie Delainey, Professor and Dean School of Nursing, University of Minnesota, USA, for reading and commenting on the manuscript. References [1] IDF. Diabetes Prevalence. 2006. International Diabetes Federation. 2006. www.idf.org/home/index.cfm?mode=264 13th September 2006. [2] UK Prospective Diabetes Study (UKPDS). Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). Lancet 1998;352:832–53. [3] Deakin T, McShane C, Cade JE, Williams R. Group based self-management education in adults with type 2 diabetes mellitus. Cochrane systematic review. Diabetes 2005;53:A515. [4] Norris SL, Engelgau MM, Narayan KM. Effectiveness of self-management training in type 2 diabetes: a systematic review of randomized controlled trials. Diabetes Care 2001;24:561–87. [5] Bruttomesso D, Gagnayre R, Leclercq D, Crazzolara D, Busata E, d’Ivernois JF, Casiglia E, Tiengo A, Baritussio A. The use of degrees of certainty to evaluate knowledge. Patient Educ Couns 2003;51:29–37. [6] Norris SL, Lau J, Smith SJ, Schmid CH, Engelgau MM. Self-management education for adults with type 2 diabetes: a meta-analysis of the effect on glycemic control. Diabetes Care 2002;25:1159–71. [7] Knight KM, Dornan T, Bundy C. The diabetes educator: trying hard, but must concentrate more on behaviour. Diabetic Med 2006;23:485–91. [8] Bodenheimer T, Lorig K, Holman H, Grumbach K. Patient self-management of chronic disease in primary care. JAMA 2002;288:2469–74. [9] Johnston-Brooks CH, Lewis MA, Garg S. Self-efficacy impacts self-care and HbA1c in young adults with Type I diabetes. Psychosom Med 2002;64:43–51. [10] Bandura A. Self-efficacy: toward a unifying theory of behavioral change. Psychol Rev 1977;84:191–215. [11] Brown SA. Interventions to promote diabetes self-management: state of the science. Diabetes Educator 1999;S52–61. [12] Steed L, Cooke D, Newman S. A systematic review of psychosocial outcomes following education, self-management and psychological interventions in diabetes mellitus. Patient Educ Couns 2003;51:5–15. [13] Norris SL, Nichols PJ, Caspersen CJ, Glasgow RE, Engelgau MM, Jack L, Isham G, Snyder SR, Carande-Kulis VG. The effectiveness of disease and case management for people with diabetes: a systematic review. Am J Prev Med 2002;22:S15–38. [14] Sigurdardottir AK. Self-care in diabetes: model of factors affecting selfcare. J Clin Nurs 2005;14:301–14.
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31 [15] Ellis SE, Speroff T, Dittus RS, Brown A, Pichert JW, Elasy TA. Diabetes patient education: a meta-analysis and meta-regression. Patient Educ Couns 2004;52:97–105. [16] Witten IH, Frank E. Data mining: practical machine learning tools and techniques. London: Elsevier; 2005. [17] Introduction to data mining and knowledge discovery. Potomac, USA: Two Crows Corporation; 2005. [18] Fayyad U, Piatetsky-Shapiro G, Smyth P. From data mining to knowledge discovery in databases. http://www.kdnuggets.com/gpsubs/aimafg-kddoverview-1996. [19] Elasy TA, Ellis SE, Pichert JW. A taxonomy for diabetes education interventions. Patient Educ Couns 2001;43:121–7. [20] http://www.cs.waikato.ac.nz/ml/index.html 23th May 2006. [21] van Tulder M, Furlan A, Bombardier C, Bouter L. Updated method guidelines for systematic reviews in the Cochrane Collaboration back review group. Spine 2003;28:1290–9. [22] Acik Y, Bulut HY, Gulbayrak C, Ardicoglu O, Ilhan N. Effectiveness of a diabetes education and intervention program on blood glucose control for patients with type 2 diabetes in a Turkish community. Southe Asian J Trop Med Public Health 2004;35:1012–8. [23] Ho¨rnsten A, Lundman B, Stenlund H, Sandstrom H. Metabolic improvement after intervention focusing on personal understanding in type 2 diabetes. Diabetes Res Clin Pr 2005;68:65–74. [24] Pibernik-Okanovic M, Prasek M, Poljicanin-Filipovic T, Pavlic-Renar I, Metelko Z. Effects of an empowerment-based psychosocial intervention on quality of life and metabolic control in type 2 diabetic patients. Patient Educ Couns 2004;52:193–9. [25] Rickheim PL, Weaver TW, Flader JL, Kendall DM. Assessment of group versus individual diabetes education: a randomized study. Diabetes Care 2002;25:269–74. [26] Surwit RS, van Tilburg MAL, Zucker N, McCaskill CC, Parekh P, Feinglos MN, Edwards CL, Williams P, Lane JD. Stress management improves long-term glycemic control in type 2 diabetes. Diabetes Care 2002;25:30–4. [27] Brown SA, Garcia AA, Kouzekanani K, Hanis CL. Culturally competent diabetes self-management education for Mexican Americans: The Starr county border health initiative. Diabetes Care 2002;25:259–68. [28] Brown SA, Blozis SA, Kouzekanani K, Garcia AA, Winchell M, Hanis CL. Dosage effects of diabetes self-management education for Mexican Americans: The Starr county border health initiative. Diabetes Care 2005;28:527–32. [29] Anderson-Loftin W, Barnett S, Bunn P, Sullivan P, Hussey J, Tavakoli A. Soul food light. Culturally competent diabetes education. Diabetes Educ 2005;31:555–63.
31
[30] Trento M, Passera P, Bajardi M, Tomalino M, Grassi G, Borgo E, Donnola BE, Cavallo F, Bondonio P, Porta M. Lifestyle intervention by group care prevents deterioration of Type II diabetes: a 4-year randomized controlled clinical trial. Diabetologia 2002;45:1231–9. [31] Kim H, Oh J, Lee H. Effects of nurse-coordinated intervention on patients with type 2 diabetes in Korea. J Nurs Care Qual 2005;20:154–60. [32] Sone H, Katagirl A, Ishibash S, Abe R, Saito Y, Murase T, Yamashita H, Yajima Y, Ito H, Ohashi Y, Akanuma Y, Yamada N. Effects of lifestyle modifications on patients with type 2 diabetes: The Japan diabetes complications study (JDCS) study design, baseline analysis and three year-interim report. Horm Metab Res 2002;34:509–15. [33] Cooper HC, Booth K, Gill G. Patients’ perspectives on diabetes health care education. Health Educ Res 2003;18:191–6. [34] Goudswaard AN, Stolk RP, Zuithoff NPA, de Valk HW, Rutten GEHM. Long-term effects of self-management education for patients with Type 2 diabetes taking maximal oral hypoglycaemic therapy: a randomized trial in primary care. Diabetic Med 2004;21:491–6. [35] Gaede P, Beck M, Vedel P, Pedersen O. Limited impact of lifestyle education in patients with Type 2 diabetes mellitus and microalbuminuria: results from a randomized intervention study. Diabetic Med 2001;18:104–8. [36] Miller CK, Edwards L, Kissling G, Sanville L. Nutrition education improves metabolic outcomes among older adults with diabetes mellitus: results from a randomized controlled trial. Prev Med 2002;34:252–9. [37] Schwedes U, Siebolds M, Mertes G. Meal-related structured self-monitoring of blood glucose: effect on diabetes control in non-insulin-treated type 2 diabetic patients. Diabetes Care 2002;25:1928–32. [38] Sarkadi A, Rosenqvist U. Experience-based group education in Type 2 diabetes: a randomised controlled trial. Patient Educ Couns 2004;53: 291–8. [39] Wolf AM, Conaway MR, Crowther JQ, Hazen KY, Nadler JL, Oneda B, Bovbjerg VE. Translating lifestyle intervention to practice in obese patients with type 2 diabetes: improving control with activity and nutrition (ICAN) study. Diabetes Care 2004;27:1570–6. [40] Deakin TA, Cade JE, Williamst R, Greenwood DC. Structured patient education: the diabetes x-pert programme makes a difference. Diabetic Med 2006;23:944–54. [41] Chodosh J, Morton SC, Mojica W, Maglione M, Suttorp MJ, Hilton L, Rhodes S, Shekelle P. Meta-analysis: chronic disease self-management programs for older adults. Ann Intern Med 2005;143:427–38. [42] Mu¨hlhauser I, Berger M. Patient education—evaluation of a complex intervention. Diabetologia 2002;45:1723–33. [43] Lauver DR, Ward SE, Heidrich SM, Keller ML, Bowers BJ, Brennan PF, Kirchhoff KT, Wells TJ. Patient-centered interventions. Res Nurs Health 2002;25:246–55.
Review
Outcomes of educational interventions in type 2 diabetes: WEKA data-mining analysis Arun K. Sigurdardottir a,*, Helga Jonsdottir b, Rafn Benediktsson c,d,e a
Faculty of Health Sciences, University of Akureyri, 600 Akureyri, Iceland b Faculty of Nursing, University of Iceland, Iceland c Department of Endocrinology and Metabolism, Landspitali University Hospital, Iceland d Faculty of Medicine, University of Iceland, Iceland e Icelandic Heart Association Iceland, Iceland Received 30 November 2006; received in revised form 4 March 2007; accepted 6 March 2007
Abstract Objective: To analyze which factors contribute to improvement in glycemic control in educational interventions in type 2 diabetes reported in randomized controlled trials (RCT) published in 2001–2005. Methods: Papers were extracted from Medline and Scopus using educational intervention and adults with type 2 diabetes as keywords. Inclusion criteria were RCT design. Data were analyzed with a data-mining program. Results: Of 464 titles extracted, 21 articles reporting 18 studies met the inclusion criteria. Data mining showed that for initial glycosylated hemoglobin (HbA1c) level 7.9% the diabetes education intervention achieved a small change in HbA1c level, or from +0.1 to 0.7%. For initial HbA1c 8.0%, a significant drop in HbA1c level of 0.8–2.5% was found. Data mining indicated that duration, educational content and intensity of education did not predict changes in HbA1c levels. Conclusion: Initial HbA1c level is the single most important factor affecting improvements in glycemic control in response to patient education. Data mining is an appropriate and sufficiently sensitive method to analyze outcomes of educational interventions. Diversity in conceptualization of interventions and diversity of instruments used for outcome measurements could have hampered actual discovery of effective educational practices. Practice implications: Participation in educational interventions generally seems to benefit people with type 2 diabetes. Use of standardized instruments is encouraged as it gives better opportunities to identify conclusive results with consequent development of clinical guidelines. # 2007 Elsevier Ireland Ltd. All rights reserved. Keywords: Educational interventions; Type 2 diabetes; Randomized controlled trials
1. Introduction Prevalence of type 2 diabetes is increasing [1]. It is a costly condition which causes significant morbidity and mortality. Good metabolic control clearly reduces the occurrence of complications [2]. Metabolic control rests heavily on self-care of the individuals having diabetes as well as their families. Since knowledge is a prerequisite for effective self-care, diabetes education has become an integral part of diabetes care [3]. The goal of the education is to enhance diabetes related self-care that contributes to good metabolic control and well-
* Corresponding author. Tel.: +354 460 8464; fax: +354 460 9999. E-mail address: [email protected] (A.K. Sigurdardottir). 0738-3991/$ – see front matter # 2007 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.pec.2007.03.007
being, which hopefully minimizes the occurrence of both acute and chronic complications. Several studies have been carried out on diabetes educational interventions. However, standardization of outcome measures is in its infancy. In a systematic review, Norris et al. [4] reported that ‘‘knowledge was measured via a variety of instruments often lacking in reported reliability and validity’’ and it seems that incorrect information or knowledge regarding diabetes treatment is common among people with diabetes [5]. Systematic reviews have indicated that diabetes education, at least in the short term, increases knowledge among people with type 2 diabetes [3,4]. Deakin et al. [3] documented in a metaanalysis that group based diabetes educational programs improved diabetes knowledge for up to 14 months. Educational interventions seem to enhance positive changes in self-reported
22
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
diet such as carbohydrate and fat intake [4]. The effectiveness of education with regard to metabolic control among adults with type 2 diabetes has been confirmed in at least two systematic reviews [6,7], where the impact of education has lowered glycosylated hemoglobin (HbA1c) by about 0.3–1.4% for as long as 12–14 months following educational intervention [3,6]. Interventions teaching self-care skills seem more effective than interventions providing information only in improving clinical outcomes [4,8]. Specific self-care instructions have been shown both to increase self-efficacy and self-care where increased self-efficacy contributes to increased self-care [9]. Self-efficacy is task specific [10] and Brown [11] maintained that the focus in diabetes education should be on narrow dietary goals and exercise as it has larger effects on weight loss and HbA1c level than general broader aimed interventions. This postulation is supported by an analysis of 36 studies on diabetes education [12] where self-care skills interventions were more frequently associated with improvements in metabolic control than informational or educational interventions. Psychosocial factors such as attitude and distress affect self-care and can modify self-care [13,14]. Diabetes educational programs have been criticized for being centered on knowledge and physiological outcomes, whereas the patient’s experience of living with the disease has not been given due concern [13]. Similarly, in Norris et al. [4] meta-analysis only 5 studies out of 72 included psychosocial variables as outcome variables. Although evidence exists regarding the ability of educational interventions to improve both knowledge and metabolic control, it is still unclear what contributes directly to this improvement. By knowing which factors are most important, health care practitioners would be better able to focus their work, improve efficiency and keep costs associated with the educational programs at a minimum. The purpose of this systematic review was to analyze educational interventions conducted in published randomized control trials (RCT) on people with type 2 diabetes. We have systematically looked for conceptual models on which interventions were based on, the delivery and teaching methods, settings, content and intensity of interventions and characteristics of providers. Measurements of outcomes were also considered. Key factors influencing improvements in glycemic control were identified by using data-mining methods. 2. Methods 2.1. Databases searched for publications The searched databases were Medline, Scopus and the Cumulative Index to Nursing and Allied Health Literature (CINAHL), including year 2001 to October week 3, 2005. Previously, Norris et al. [4,6] and Ellis et al. [15] analyzed RCT studies published up to the year 2000. A librarian assisted with the database search. Terms searched were diabetes mellitus.mp or exp *Diabetes Mellitus, patient education/or educational intervention.mp, including all subheadings. The terms were also combined and limited to the English language and adults with type 2 diabetes.
Exclusion criteria were set as; diabetes type 1 or if type of diabetes was not reported in the study or if results were not analyzed according to type of diabetes. Studies regarding gestational diabetes, pilot studies and preliminary reports were excluded. Titles of articles were independently analyzed by two persons in order to ensure that relevant studies were retrieved. Reference lists of retrieved articles were also scrutinized. 2.2. Data mining Data mining and knowledge discovery in databases (KDD) was used to analyze the data. The KDD process maps low-level data that are often too voluminous to be understood into forms that are more compact or useful in business [16] or even health care. The KDD process uses specific data-mining methods (computer algorithms) for pattern discovery and extraction. Data mining involves finding patterns or rules which represent new knowledge or ideas which then can be used to make predictions or form the basis of hypotheses for future experiments [16–18]. Computer algorithms are used to find rules or decision trees that best explain a data set. A branch in a rule or decision tree occurs for particular values of an attribute (variable) and the consequent rule or the leaf nodes of a decision tree, makes the final classification for the attribute chosen as the attribute being predicted or the dependent attribute. Computer algorithms used to find rules or decision trees work by recursively selecting the attribute which does the best job of classifying (discriminating) the attribute being predicted [16,18]. This tactic is employed because it usually results in rules or decision trees which are quite small and which can be used to interpret the data set. Typically few of the attributes in the data set are selected to form the rule or decision tree [16]. A data set to be mined is described by attributes (variables) and instances in this case a single study. Numeric and nominal attributes are used to describe a data set to be mined. Nominal attributes take on values in a pre-specified, finite set of possibilities [16]. In our data set, we converted all numeric attributes into nominal attributes, see Section 2.5. Typically, one of the attributes is taken as the dependent attribute representing the concept to be predicted by a pattern or a rule [16,17]. However, it is possible to consider each attribute as the dependent attribute. Coding attributes and checking them for incorrect attribute values and typographic errors is important. Results of initial data-mining attempts can suggest that attributes should be removed or combined; such actions require that the data set is carefully prepared before the subsequent KDD process [16–18]. An instance is characterized by values of the attributes. In the data set analyzed in this paper, an instance characterizes a single study. When using data mining to predict models or rules it is important that instances are correctly predicted, or the accuracy of the prediction is high. Accuracy is the fraction of examples predicted correctly by data mining according to the model and measures how error-free the model’s predictions are [17]. Error in the model prediction is often referred to as a misclassification [16,17].
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
2.3. Data extraction When extracting factors to use in data mining, Elasy et al.’s [19] taxonomy regarding educational interventions was applied for the intervention group, with minor additions, such as whether a theory was guiding the intervention and mean age of study participants. Intensity of the intervention was calculated as the number of episodes multiplied by duration of episodes to obtain how many hours the intervention lasted. Duration of intervention in months was a separate variable in data mining. When multiple post-intervention values (glycosylated hemoglobin (HbA1c) level, body mass index (BMI)) were available the newest value was used. Difference in HbA1c level pre- and post-intervention (Idiff) is the main outcome analyzed by data mining. However, other variables were also tested as dependent attributes, including theory, provider (nurse, physician and research team), intensity of the intervention, duration of the intervention, follow-up time, follow-up time after the intervention finished, mean age and reduction in BMI. 2.4. The machine learning program WEKA Waikato Environment for Knowledge Analysis or WEKA, Version 3.4.3 was used for data mining. WEKA was developed at the University of Waikato in New Zealand [16]. The WEKA program is written in Java, is available on the Internet [20], and comprises a variety of data-mining algorithms. In this study, we made use of the J48 classifier algorithm which is an implementation of the C4.5 decision tree learner, see Witten and Frank p. 198 [16]. 2.5. Data analysis Initial data-mining attempts involved considering which attributes to code and how they should be coded by studying histograms of values and doing many data-mining attempts. Histograms of each attribute were examined to determine natural groupings of values into categories as well as finding mean values of attributes. Cutoff points occurred at natural dips in the histograms. Since data-mining attempts with three or four categories returned large rules with little meaning, the final attempt simply used two possible categories for each attribute. Using two possible categories for every attribute in this way also eliminated bias in the data-mining process towards attributes having many possible categories over attributes with few possible values. The best definitions were set as follows: The HbA1c level was coded as: The attributes Idiff and Cdiff represented the difference in HbA1c level between pre- and post-intervention for the intervention and control groups, respectively. Idiff had two possible values defined as: (a) small change, +0.1 to 0.7% (9 studies), (b) big reduction, from 0.8 to 2.5% (9 studies). Cdiff also had two possible values defined as: (a) small change from +1.2 to 0.4% (12 studies) and (b) big reduction, from 0.5 to 1.7% (5 studies).
23
Initial HbA1c level was coded as: high, 8.0% (8 studies) and low 7.9% (10 studies). Intensity of intervention was coded as: (a) less, <11 h (6 studies) and (b) more, 11 h (12 studies). Duration of intervention was defined as: (a) short, <7 months (10 studies) and (b) long, 7 months (7 studies). The follow-up assessment time was defined as: (a) less, <12 months (7 studies) and (b) more, 12 months (11 studies). Attributes representing setting, delivery, teaching method, provider and content were given yes (y) or no (n) definitions, e.g. if diet was addressed in intervention then it was classified with y. A t-test was also used to test for difference in HbA1c level between groups, to supplement the data-mining results. 3. Results The search of databases resulted in 464 publications. From Medline, 19 articles were retrieved reporting 16 studies. From Scopus, one study not reported in Medline and one study from reference lists also not reported in Medline met our inclusion criteria. As CINAHL was the last database to be searched, all articles (n = 17) reported in it had already been retrieved from the other databases. The total numbers of included studies were 18, reported in 21 articles. Fig. 1 shows the systematic review flow diagram and reasons for exclusion. 3.1. Validity assessment Internal validity was assessed by van Tulder’s et al. [21] scheme for assessment of internal validity, where high quality of a research is higher or equal to 6 on a scale from 0 to 11 [21]. Two reviewers independently assessed the internal validity of the interventions. The assessments were compared and agreed upon. In the validity assessment, 7 studies of 18 received scoring of 6 or higher, or good scoring and two studies received scoring of one, see Table 1. One study, Acik et al. [22], used three groups. Of those we excluded one group in the analysis (WEKA), the diet-exercise group, as selection into it was based on extremely high pre HbA1c level. 3.2. Educational interventions Fig. 2 shows the taxonomy of the interventions, e.g. how often theory is guiding the intervention or what is included in the interventions’ content. The taxonomy for each intervention is not mutually exclusive as for example, teaching methods for one intervention could include all four teaching styles. Table 2 is a summary table of the interventions and the main findings displayed in a chronological order. 3.3. Theory guided interventions In ten cases of 18 a theoretical framework guiding the intervention was reported. Within the theoretical frameworks
24
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
Fig. 1. Systematic review flow diagram of retrieved studies, n, number of articles.
there existed diversity. For example, Ho¨rnsten et al. [23] evaluated whether person centered intervention was more effective than conventional diabetes care. An empowerment theory guided Pibernik-Okanovi et al.’s [24] and Rickheim et al.’s [25] studies. Culturally appropriate educational
programs in minority populations in USA were the focus of four studies concerning Native Americans [26], Mexican Americans [27,28] and African–Americans [29]. When testing whether a theoretical framework affected how the intervention was organized (content), how long the
Table 1 Assessment of methodological quality; highest possible score is 11 and the lowest is 0 Reference
Acik et al. [22] 1
AndersonLoftin [29] 7
Brown et al. [28] 6
Brown et al. [27] 6
Cooper et al. [33] 4
Gaede et al. [35] 5
Reference
Miller et al. [36]
Rickheim et al. [25]
Sarkadi and Rosen-qvist [38]
Schwedes et al. [37]
Sone et al. [32]
Total quality score
4
Pibernik Okanovic et al. [24] 1
4
4
5
6
Total quality score
Ho¨rnsten et al. [23] 4
Kim et al. [31] 3
Surwit et al. [26]
Trento et al. [30]
Wolf et al. [39]
2
8
9
Goudswaard et al. [34] 8
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
25
Fig. 2. Percentage of interventions (n = 18) using each attribute in the intervention. The y-axis represents % of studies including exact attribute, where 18 = 100%. The x-axis shows attributes such as setting, provider, teaching and delivery methods and content of the interventions.
follow-up time was or difference in BMI and HbA1c level preto post-intervention no influence was found in data mining. No statistically significant difference in reduction in HbA1c level was found between theory guided interventions and non-theory guided interventions (t(16) = 0.66, p = 0.516). 3.4. Setting, delivery, provider and teaching method Group education in combination with individual education was commonly used as 13 of 18 interventions applied that form. Two studies compared group and individual education [25,30]. The most common delivery method was face to face (17 studies) and one study [31] used a combination of telecommunication and written instructions. Sone et al. [32] used three types of delivery, face to face, telecommunication and written instructions. Providers of interventions were nurses [23,33] physicians [22,30], team of health care providers [27,28,34,35] and dietitians [36]. All interventions apart from one [22] used collaborative teaching methods such as goal setting, problem solving and cognitive reframing. Differences in setting, delivery, provider and teaching method were not related to differences in HbA1c level pre- to post-intervention. 3.5. Content of interventions Content was most often reported as teaching about basic diabetes knowledge and self-care skills such as diet and exercising, medication adherence, self-monitoring of blood glucose (SMBG) and psychosocial aspects. Three studies explicitly declared that education was based on initial assessment [24,36,37]. However, it is unclear from two papers [24,37] how initial assessment was conducted and whether interventions were tailored to the assessment.
3.5.1. Knowledge Although almost all interventions included basic diabetes knowledge, measuring knowledge as an outcome was only conducted in five studies using different instruments. Brown et al. [27,28] and Trento et al. [30] tested knowledge as an outcome with instruments that they had developed but Sarkadi and Rosenqvist [38] did not detail how they measured knowledge. All the above-mentioned studies demonstrated improvements in knowledge with education, even four years after the intervention [30]. When testing if measurement of knowledge affected reduction in HbA1c level post-intervention, data mining did not detect difference and no significant difference was found (t(16) = 0.03, p = 0.490). 3.5.2. Self-care skills Self-care skills were classified into four areas as: diet or nutrition, physical activity, medication adherence and SMBG [14]. Most of the interventions included self-care areas and ten interventions described all these four self-care areas in their educational content. However, self-care skills were seldom measured and the measurements that were carried out included diverse instruments. Cooper et al. [33] measured self-care by the Diabetes Self-Care questionnaire and Anderson-Loftin et al. [29] measured dietary fat behavior by the FHQ-25 questionnaire. SMBG measurements were included in nine interventions, see Table 2. Comparison of groups that included SMBG in the intervention to those that did not demonstrated no difference in HbA1c reduction according to data mining and ttest pre- to post-intervention (t(9) = 1.72, p = 0.119). 3.5.3. Emotional aspects Of 13 studies which included teaching emotional aspects, only eight interventions measured the effect of this inclusion.
26
Table 2 Characteristics and outcomes of educational interventions included in the review Focus of paper, participants
Intervention
Outcomes
Conclusion
Gaede et al., Danmark [35]
To assess the effect of intensified education, on diet, exercise, and smoking (I: 80, C: 80)
All patients in both groups informed of basic diabetes guidelines, I: individual diet interventions and patients set their own goals. Group intervention regarding diet, exercise and smoking, C: nr (I I: more, D I: more, F: more)
Metabolic control + I, metabolic control C, dietary behaviour+, weight reduction I and C, no difference between groups in time used to exercise
Dietary behavior analyzed other self-care skills not measured. High resources and only modest changes in behavior
Miller et al., USA [36]
To evaluate nutrition intervention on blood glucose and lipoprotein levels (I: 47, C: 45)
I: 10-w sessions in healthy eating. Each session lasted 90 min. Theory of Meaningful learning and Social Cognitive Theory, C: usual care (I I: more, D I: less, F: less)
Metabolic control + I, metabolic control, nc C, treatment goals for total cholesterol+
Nutrition education improves metabolic control among older adults
Schwedes et al., Germany and Austria [37]
To investigate the effect of meal-related SMBG on glycemic control and well-being in non-insulin treated people (I: 113, C: 110)
I: instruction in use of SMBG device and measure BG six times daily, before and 1 h after main meal and record the values in a diary for BG and well-being. Seen by a nurse every 4 w, six times and asked to reflect on self-regulation and life with diabetes, C: counseling on diet and lifestyle (I I: more, D I: less, F: less)
Metabolic control + I, metabolic control + C, weight reduction + I, weight reduction + C, satisfaction with treatment I + and C, well-being + I
Keeping a diary and recording eating habits and SMBG enabled more autonomous self-care
Brown et al., USA [27]
To compare 2 interventions designed for Mexican Americans (I: 126, C: 114)
I: 2-h weekly sessions for 3-m, diet, exercising, SMBG. 6-m of 2-h biweekly and 3-m of monthly 2-h sessions to promote behavior change. 52-h contact. Family member brought to the program., C: usual care, 1-y waiting list (I I: more, D I: more, F: more)
Metabolic control + I, metabolic control C, BMI + I, BMI C, knowledge + I and C
Culturally specific education for Mexican Americans is beneficial. High resources and only modest changes in behavior
Surwit et al., USA [26]
To assess if stress management can improve glucose control (I: 60, C: 48)
C: five 30 min sessions on exercise and diet, and videos featuring how Native Americans should incorporate traditional Native American values and foods into life, I: in addition to the same education as C, there was a stress management program (I I: more, D I: less, F: more)
Metabolic control + I, metabolic control C, stress-management + I group
Intervention supports efficacy of stress management training. Self-care skills not measured
Trento et al., Italy [30]
To compared traditional individual diabetes care with a model in which routine follow-up is managed by interactive group visits (I: 56, C: 56)
I: group education very 3-m, diet, exercise, SMBG, medication and complications. Annually diabetes screening, C: visit every 3-m to same physician, knowledge assessed and education offered, diet, SMBG, exercising (I I: more, D I: more, F: more)
Metabolic control + I, metabolic control C, BMI + I, BMI + C, quality of life + I, knowledge + I and C, self-care+
Group care is feasible in an ordinary clinic and costeffective in preventing the deterioration of metabolic control and quality of life
Sone et al., Japan [32]
To asses if long term life-style intervention can improve glycemic control and prevent complications (I: 1105, C: 1100)
I: dietary habits, physical activities and adherence to treatment. 15 min telephone counseling sessions at least once every two weeks, C: usual care (I I: more, D I: nr, F: more)
Metabolic control + I and C, BMI + I, BMI nc C
No instruments used to measure QoL, well-being or self-care skills
Rickheim et al., USA [25]
To compare effectiveness of delivering diabetes education in either a group or individual setting using consistent evidence-based curriculum. (I: 87 (group education) C: 83 (individual education))
Both groups received the same education, diet, SMBG, physical activity, foot care, problem solving emphasizing to empower and meeting the needs of an adult learner, 5–7 h of education over 6-m (I I: less, D I: less, F: less)
Metabolic control + I and C, BMI + I and C, knowledge + I and C, attitude + I and C, QoL + I and C
Group education is as effective as individual education. The optimal group size is not known
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
Study
To find if intervention impacting upon illness beliefs and leading to changes in self-care behaviors affects blood glucose control (I: 53, C: 36)
8 weekly group sessions, 2-h each, diabetes nurses, experiential learning, diet, exercise, emotions, SMBG goals, behavior change, C: nr (I I: more, D I: less, F: more)
Metabolic control I, metabolic control + C, attitude + I, self-care + I and C
Reflection was useful to clarify and interpret complexities of self-care. Supportive environment was helpful and patients valued collaborative learning
Wolf et al., USA [39]
To compare efficacy of lifestyle education to usual care in primary care (I: 73, C: 74)
I: case manager met patients individually, in groups and by phone. Education, support and goal setting. Goals tailored to patients but met dietary recommendations for people with type 2 diabetes and obesity, C: usual care (I I: less, D I: more, F: more)
Metabolic control + I, metabolic control C, weight reduction + I, weight reduction C, emotional aspects + I
Change in weight did not predict change in HbA1c level. Weight gain during the last 4 months suggests need for ongoing lifestyle coaching. Self-care skills not measured
Gouds-waard et al., Netherlands [34]
To assess the long term effects of education (I: 25, C: 29)
I: collaborative education, by 2 diabetes nurses, diet, physical exercising and losing body weight, medication adherence, SMBG, BG control at home. 6 sessions over 6-m, total contact time of 2.5 h, C: usual care, 3-m reviews, focusing on diabetic symptoms and measurement of BG (I I: less, D I: less, F: more)
Metabolic control + I, metabolic control + C, weight reduction + I
Long-term effects were disappointing. HbA1c level and weight loss occurred at the same time. No instruments used to measure QoL or self-care skills
Acik et al., Turkey [22]
To analyze effects of patient education, diet and regular exercise on blood glucose control (I: 33, C: 33)
I: diet counseling C: usual care (I I: less, D I: less, F: less)
Metabolic control + I, metabolic control C, BMI + I, BMI C
No instruments used to measure QoL, well-being or self-care skills
Pibernik-Okanovic et al., Croatia [24]
To determine the feasibility of empowerment educational program on patients’ quality of life and HbA1c level (I: 73, C: 35)
I: 6 w group sessions, goal setting, problem solving, coping with diabetes, seeking social support, and staying motivated. Discussion and practical exercises, C: Usual care (I I: less, D I: less, F: less)
Metabolic control + I, metabolic control, nr C, self-care behaviors + I, quality of life + I
Patients perceived benefits from the program, support from professionals, not being criticized and stimulated for self-care
Sarkadi and Rosenqvist, Sweden [38]
To present an educational program based on participants’ experiences and use these experiences as a basis for acquisition of practical skills needed for self-management of diabetes (I: 39, C: 38)
I: 12 m group education, once in a month, where practical aspects of diabetes were analyzed by the group, such as choice and preparation of food, SMBG tasks, and how exercising decreased BG and support to deal with emotional aspects of diabetes, C: usual care (I I: more, D I: more, F: more)
Metabolic control + I, metabolic control C, knowledge+, exercising + I
Self-reported chances in weight., Assessment of exercise not other self-care skills
Brown et al., USA [28]
To compare 2 interventions designed for Mexican Americans, Program was designed according to the culture (I: 102 extended, C: 114 compressed)
I: extended, 12-w, 2-h sessions, diet, exercising, SMBG, medication adherence; 14, support group sessions, goals, problem solving 2-h each, C: compressed 8-w, 2-h support group sessions, goals, problem solving and support sessions held at 3, 6 and 12 mI (I: more, D I: more, F: more)
Metabolic control + I and C, knowledge + I and C, no significant difference found between programs
The most effective dosage of education is unknown. Selfcare skills or emotional aspects not measured
Ho¨rnsten et al., Sweden [23]
To evaluate whether a person centered intervention focusing on patients’ personal understanding of their illness is more effective than is conventional diabetes care (I: 44, C: 60)
I: 10 group sessions, 2 h each, discussion not predetermined, emanated from patient understanding and covered experiences and practical theoretical points, C: regular care in other clinics, 1–2 y visits to nurses in other clinics (I I: more, D I: more, F: more)
Metabolic control + I, metabolic control C, BMI + I, BMI C, satisfaction + I, well-being nc
No self-care skills evaluated. Focus on wellness in patients’ experiences may lead to decisions to manage the disease
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
Cooper et al., UK [33]
27
, negative change; nc, no change; nr, not reported; BG, blood glucose; SMBG, self-monitoring blood glucose; I I, intensity of I, intervention; C, control; h; hour; y, year; w, week; m, month; +, positive change; intervention; D I, duration of intervention; F, follow-up assessment.
Intervention using individualized approach through telephone improves blood glucose control and satisfaction Metabolic control + I, metabolic control C, satisfaction + I I: booklet regarding diet, exercise, medication and SMBG and keeping diet and exercise diaries, hypo- and hyperglycemia management. Nurse phone calls weekly for 12-w, C: nr (I I: less, D I: less, F: less) To investigate the effect of nurse telephone calls on HbA1c levels and adherence to self-care recommendations (I: 15, C: 10) Kim et al., Korea [31]
Culture is an important part of life and affects dietary chooses. Self-care skills or emotional aspects not measured Metabolic control + I, metabolic control + C, BMI + I, BMI C, dietary behavior + I I: four weekly classes on diet, included in social events. Four monthly 1-h peerprofessionals discussion groups, learning about diabetes. Weekly follow-up calls by nurses, C: diabetes classes for 8-h (I I: more, D I: less, F: less) To test culturally appropriate dietary self-management intervention in African– Americans (I: 49, C. 48) Anderson-Loftin et al., USA [29]
Conclusion Outcomes Focus of paper, participants
Intervention
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
Study
Table 2 (Continued )
28
Results were improvements in well-being, anxiety and attitudes towards diabetes. However, different instruments were applied; the Medical Outcome Study Short Form (SF-36) and the adjustment to diabetes instrument (ATT19) [25,39], the 22 well-being questionnaire [24,37], the WHO-QoL instrument [24], the Stage-Trait Anxiety [26] and the Diabetes Integration questionnaire [33]. Using data mining to test whether educational content affected reduction in HbA1c level pre- to post-intervention, attributes such as setting, delivery, teaching methods and provider were removed from data mining, but the results were unchanged. 3.6. Intensity and duration of intervention Most programs, or 12 of 18, used more than 11 h of intervention. The duration of interventions was from 8 weeks [22] to 12 months [28,30,34,35,39] and the follow-up assessment from 8 weeks [22] to 51 months [30]. Only seven interventions had follow-up time for more than 6 months after the intervention finished [26,30,32–35,38]. The WEKA data-mining program did not detect association between reduction in HbA1c level and these attributes. No statistical significant difference was found between groups according to follow-up time for more than 6 months after the intervention finished compared to less than 6 months on HbA1c level (t(16) = 1.57, p = 0.136). 3.7. Biological factors (HbA1c level and BMI) Pre intervention HbA1c level predicted improvement in glucose control in response to patient education. If initial HbA1c level was high or 8% the reduction was 0.8–2.5% and if the initial HbA1c level was 7.9% the change was from +0.1 to 0.7%. Studies that started with low initial HbA1c level achieved less HbA1c reduction as a response to the education. Statistically significant difference was also found between intervention groups (Idiff) in HbA1c mean reduction according to high or low initial HbA1c level (t(11) = 3.53, p = 0.004). Three studies were misclassified, e.g. the intervention achieved higher or lower HbA1c reduction than predicted according to start level. The accuracy of the prediction is 83.4% [19]. Two of the misclassified studies [22,24] had low internal validity, with scoring of 1 out of 11, see Table 1. In the third misclassified study (Goudswaard et al. [34]) the participants were all treated with maximal dosages of oral hypoglycemic agents. Goudswaard et al.’s [34] results were promising at 6 weeks as the intervention group had improved their glucose control significantly compared to the control group, but at 18 months there was no significant difference in HbA1c level as regarded the number of patients with HbA1c <7.0% or the number of patients treated with insulin. To detect if studies with low internal validity scoring would affect data-mining results, studies with internal validity 5 were removed from data mining. This left 7 interventions to be mined but the results were unchanged.
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
Fig. 3. Percentage change in glycated hemoglobin – the HbA1c level – for intervention and control group. The y-axis represents % change in HbA1c level as a response to education. Time on x-axis is number of months from beginning of the intervention. The symbol * shows significant difference between intervention and control groups in HbA1c level.
For the interventions groups the relative HbA1c level reduction is on average 6–7% compared to the control groups’, see Fig. 3. There was also a statistically significant difference in HbA1c mean reduction, between the intervention and the control groups (t(33) = 2.82, p = 0.008), even though in at least seven interventions (Table 2) the control group received more care than standard care. Seven interventions out of 18 achieved more than 10% reduction in HbA1c level and four of these interventions achieved post-HbA1c level 7.0%. Seven studies reported difference in BMI but WEKA did not demonstrate predictors of change in BMI. 4. Discussion and conclusion 4.1. Discussion Using data mining, this analysis of the literature on RCT of educational interventions found that HbA1c level before participating in the educational intervention was the main factor affecting how large an effect the intervention had on glucose control in adults with type 2 diabetes. This is in agreement with several other studies [26,28,32]. For instance, Sone et al. [32] reported that decrease in HbA1c level occurred only in patients with HbA1c level 7.5% or higher regardless of intervention or control group. As data mining has only recently
29
been used to analyze educational interventions it was decided to accompany that method with a more traditional statistics (t-test) to supplement our findings. Results from data mining and t-test were congruent. Accuracy of data mining was sufficient with three misclassifications, where two of the misclassified studies had low internal validity. Threats to internal validity were common but no study fulfilled van Tulder’s et al. criteria for methodological quality [21]. However, removing studies with low internal validity from data mining did not change the results. It should though be considered that this analysis included only 18 studies. Our findings that delivery, teaching methods or content did not influence reduction in HbA1c level, is in accordance with results from Norris et al. [4] but conflicts Ellis et al. [15] results where face to face delivery, exercise in the content and cognitive reframing explained 44% of variance in glucose control. In this review all but one intervention [36] included face-to-face delivery and didactic teaching method was applied in only one intervention [22]. All the other interventions used goal setting, cognitive reframing and support as teaching methods. That conflicts with results from Knight et al. [7] who found didactic teaching methods to be most common. The explanation could be that our review included more recent studies than Knight et al’s study. In the year 2006, Deakin et al. [40] published results from a RCT expert program among people with type 2 diabetes. Their results were concurrent with our results as the expert program increased knowledge and psychosocial adjustment to diabetes and improved metabolic control among the intervention group (n = 150), the HbA1c level reduced by 0.6% but in the control group (n = 141) it increased by 0.1%. Our results that contact time did not affect the HbA1c level conflicts results from Norris et al. [6], in which 15 studies out of 31 measured contact time according to the HbA1c level. For every additional hour of contact time, the HbA1c level was reduced by 0.04% (CI 0.04–0.08). The results that knowledge and self-care skills were rarely measured are in accordance with results from Bodenheimer et al. [8]. They analyzed 72 studies of self-management training in diabetes and found that 46 studies measured effects of patient education on patient knowledge and performance of technical skills, 33 studies showed positive impact and significant differences between groups were observed in 13 studies. However, Bodenheimer et al. [8] did not define performance of technical skills and how it was measured. Half of the interventions analyzed in this paper were explicitly guided by theoretical frameworks. As the interventions were based on different theories it is difficult to compare them and decide upon which is the most effective theory to guide diabetes education, a result that has been reported before [7]. The desired amount or intensity of education is unknown [6,8,28,41], which is supported by the fact that no difference was found in the post-intervention HbA1c level according to type or content of intervention, intensity, duration or length of intervention. Diversity of educational interventions in this review could have hampered
30
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31
actual knowledge discovery in databases (KDD) of effective educational practices, which is in agreement with conclusions of other authors [6,8,15,41]. In this review, seven interventions out of 18 achieved more than 10% relative reduction in HbA1c level and the interventions groups’ HbA1c level was on average 6–7% lower than the control groups’. The control groups frequently received better care than standard care which can underestimate effects of the interventions. This is an issue that has been reported before [7] and needs to be highlighted. The studies in this review included mostly people who had been diagnosed with diabetes for longer than one year. Improvement in metabolic control in spite of the fact that the patients were well beyond the time of becoming diagnosed with the diabetes and the potential motivation which that may hold is encouraging and of particular clinical significance. Some studies showed that HbA1c level [34] increased with longer follow-up time. This may indicate that people with diabetes might benefit from periodic reinforcement of education and support even throughout life. More research is warranted in this area and also on how time from diabetes diagnosis affects result of educational interventions. Individuals in the educational programs seemed to benefit from participation, shown by increased knowledge, self-care skills and improvements in psychological aspects. However, these aspects were measured with heterogeneous instruments which make comparison difficult. Therefore the effectiveness of these aspects is not yet fully clear. This analysis included only RCT studies. Mu¨hlhauser and Berger [42] caution that systematic reviews using RCTs might exclude some studies that are beneficial, although systematic reviews are in general valuable. It is known that RCT study design is not always feasible when researching factors affecting self-care behavior among people with diabetes. Interventions intended to change behavior should focus on increased knowledge and ability of individuals to utilize the knowledge in their self-care and they need to know that psychosocial aspects can influence self-care. Such interventions have to be patient centered and tailored to the patient’s needs. Tailored patient-centered interventions can only partly be standardized [43] which makes utilization of RCT more complex and even requires other methodological approaches. 4.2. Conclusion Data mining is an appropriate and sensitive method for analyzing educational interventions. These results revealed that there exists incongruity in designing and studying educational interventions which results in difficulties when considering the best intervention to use for motivating self-care behavior among people with type 2 diabetes. Usual care was often inadequately defined and sometimes the control group was offered education that could mask the differences between the groups. The instruments used to measure outcomes were various and highlight the need for universal use of standardized validated instruments.
4.3. Practice implications Participation in educational interventions is generally beneficial for people with type 2 diabetes. People with type 2 diabetes might need support and education throughout life as effectiveness of interventions seems to lessen as time passes. People with type 2 diabetes might benefit from frequent health provider contact but the intensity does not necessarily need to be so high. We encourage development of patient centered interventions and development and use of standardized validated instruments which might facilitate identification of the best method for diabetes education and consequently the introduction of universal clinical guidelines. Acknowledgements Assistance from Dr. Andrew Brooks MACM MIEEE MBCS, Associate Professor University of Akureyri Iceland, with data mining is acknowledged. We like to thank Connie Delainey, Professor and Dean School of Nursing, University of Minnesota, USA, for reading and commenting on the manuscript. References [1] IDF. Diabetes Prevalence. 2006. International Diabetes Federation. 2006. www.idf.org/home/index.cfm?mode=264 13th September 2006. [2] UK Prospective Diabetes Study (UKPDS). Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). Lancet 1998;352:832–53. [3] Deakin T, McShane C, Cade JE, Williams R. Group based self-management education in adults with type 2 diabetes mellitus. Cochrane systematic review. Diabetes 2005;53:A515. [4] Norris SL, Engelgau MM, Narayan KM. Effectiveness of self-management training in type 2 diabetes: a systematic review of randomized controlled trials. Diabetes Care 2001;24:561–87. [5] Bruttomesso D, Gagnayre R, Leclercq D, Crazzolara D, Busata E, d’Ivernois JF, Casiglia E, Tiengo A, Baritussio A. The use of degrees of certainty to evaluate knowledge. Patient Educ Couns 2003;51:29–37. [6] Norris SL, Lau J, Smith SJ, Schmid CH, Engelgau MM. Self-management education for adults with type 2 diabetes: a meta-analysis of the effect on glycemic control. Diabetes Care 2002;25:1159–71. [7] Knight KM, Dornan T, Bundy C. The diabetes educator: trying hard, but must concentrate more on behaviour. Diabetic Med 2006;23:485–91. [8] Bodenheimer T, Lorig K, Holman H, Grumbach K. Patient self-management of chronic disease in primary care. JAMA 2002;288:2469–74. [9] Johnston-Brooks CH, Lewis MA, Garg S. Self-efficacy impacts self-care and HbA1c in young adults with Type I diabetes. Psychosom Med 2002;64:43–51. [10] Bandura A. Self-efficacy: toward a unifying theory of behavioral change. Psychol Rev 1977;84:191–215. [11] Brown SA. Interventions to promote diabetes self-management: state of the science. Diabetes Educator 1999;S52–61. [12] Steed L, Cooke D, Newman S. A systematic review of psychosocial outcomes following education, self-management and psychological interventions in diabetes mellitus. Patient Educ Couns 2003;51:5–15. [13] Norris SL, Nichols PJ, Caspersen CJ, Glasgow RE, Engelgau MM, Jack L, Isham G, Snyder SR, Carande-Kulis VG. The effectiveness of disease and case management for people with diabetes: a systematic review. Am J Prev Med 2002;22:S15–38. [14] Sigurdardottir AK. Self-care in diabetes: model of factors affecting selfcare. J Clin Nurs 2005;14:301–14.
A.K. Sigurdardottir et al. / Patient Education and Counseling 67 (2007) 21–31 [15] Ellis SE, Speroff T, Dittus RS, Brown A, Pichert JW, Elasy TA. Diabetes patient education: a meta-analysis and meta-regression. Patient Educ Couns 2004;52:97–105. [16] Witten IH, Frank E. Data mining: practical machine learning tools and techniques. London: Elsevier; 2005. [17] Introduction to data mining and knowledge discovery. Potomac, USA: Two Crows Corporation; 2005. [18] Fayyad U, Piatetsky-Shapiro G, Smyth P. From data mining to knowledge discovery in databases. http://www.kdnuggets.com/gpsubs/aimafg-kddoverview-1996. [19] Elasy TA, Ellis SE, Pichert JW. A taxonomy for diabetes education interventions. Patient Educ Couns 2001;43:121–7. [20] http://www.cs.waikato.ac.nz/ml/index.html 23th May 2006. [21] van Tulder M, Furlan A, Bombardier C, Bouter L. Updated method guidelines for systematic reviews in the Cochrane Collaboration back review group. Spine 2003;28:1290–9. [22] Acik Y, Bulut HY, Gulbayrak C, Ardicoglu O, Ilhan N. Effectiveness of a diabetes education and intervention program on blood glucose control for patients with type 2 diabetes in a Turkish community. Southe Asian J Trop Med Public Health 2004;35:1012–8. [23] Ho¨rnsten A, Lundman B, Stenlund H, Sandstrom H. Metabolic improvement after intervention focusing on personal understanding in type 2 diabetes. Diabetes Res Clin Pr 2005;68:65–74. [24] Pibernik-Okanovic M, Prasek M, Poljicanin-Filipovic T, Pavlic-Renar I, Metelko Z. Effects of an empowerment-based psychosocial intervention on quality of life and metabolic control in type 2 diabetic patients. Patient Educ Couns 2004;52:193–9. [25] Rickheim PL, Weaver TW, Flader JL, Kendall DM. Assessment of group versus individual diabetes education: a randomized study. Diabetes Care 2002;25:269–74. [26] Surwit RS, van Tilburg MAL, Zucker N, McCaskill CC, Parekh P, Feinglos MN, Edwards CL, Williams P, Lane JD. Stress management improves long-term glycemic control in type 2 diabetes. Diabetes Care 2002;25:30–4. [27] Brown SA, Garcia AA, Kouzekanani K, Hanis CL. Culturally competent diabetes self-management education for Mexican Americans: The Starr county border health initiative. Diabetes Care 2002;25:259–68. [28] Brown SA, Blozis SA, Kouzekanani K, Garcia AA, Winchell M, Hanis CL. Dosage effects of diabetes self-management education for Mexican Americans: The Starr county border health initiative. Diabetes Care 2005;28:527–32. [29] Anderson-Loftin W, Barnett S, Bunn P, Sullivan P, Hussey J, Tavakoli A. Soul food light. Culturally competent diabetes education. Diabetes Educ 2005;31:555–63.
31
[30] Trento M, Passera P, Bajardi M, Tomalino M, Grassi G, Borgo E, Donnola BE, Cavallo F, Bondonio P, Porta M. Lifestyle intervention by group care prevents deterioration of Type II diabetes: a 4-year randomized controlled clinical trial. Diabetologia 2002;45:1231–9. [31] Kim H, Oh J, Lee H. Effects of nurse-coordinated intervention on patients with type 2 diabetes in Korea. J Nurs Care Qual 2005;20:154–60. [32] Sone H, Katagirl A, Ishibash S, Abe R, Saito Y, Murase T, Yamashita H, Yajima Y, Ito H, Ohashi Y, Akanuma Y, Yamada N. Effects of lifestyle modifications on patients with type 2 diabetes: The Japan diabetes complications study (JDCS) study design, baseline analysis and three year-interim report. Horm Metab Res 2002;34:509–15. [33] Cooper HC, Booth K, Gill G. Patients’ perspectives on diabetes health care education. Health Educ Res 2003;18:191–6. [34] Goudswaard AN, Stolk RP, Zuithoff NPA, de Valk HW, Rutten GEHM. Long-term effects of self-management education for patients with Type 2 diabetes taking maximal oral hypoglycaemic therapy: a randomized trial in primary care. Diabetic Med 2004;21:491–6. [35] Gaede P, Beck M, Vedel P, Pedersen O. Limited impact of lifestyle education in patients with Type 2 diabetes mellitus and microalbuminuria: results from a randomized intervention study. Diabetic Med 2001;18:104–8. [36] Miller CK, Edwards L, Kissling G, Sanville L. Nutrition education improves metabolic outcomes among older adults with diabetes mellitus: results from a randomized controlled trial. Prev Med 2002;34:252–9. [37] Schwedes U, Siebolds M, Mertes G. Meal-related structured self-monitoring of blood glucose: effect on diabetes control in non-insulin-treated type 2 diabetic patients. Diabetes Care 2002;25:1928–32. [38] Sarkadi A, Rosenqvist U. Experience-based group education in Type 2 diabetes: a randomised controlled trial. Patient Educ Couns 2004;53: 291–8. [39] Wolf AM, Conaway MR, Crowther JQ, Hazen KY, Nadler JL, Oneda B, Bovbjerg VE. Translating lifestyle intervention to practice in obese patients with type 2 diabetes: improving control with activity and nutrition (ICAN) study. Diabetes Care 2004;27:1570–6. [40] Deakin TA, Cade JE, Williamst R, Greenwood DC. Structured patient education: the diabetes x-pert programme makes a difference. Diabetic Med 2006;23:944–54. [41] Chodosh J, Morton SC, Mojica W, Maglione M, Suttorp MJ, Hilton L, Rhodes S, Shekelle P. Meta-analysis: chronic disease self-management programs for older adults. Ann Intern Med 2005;143:427–38. [42] Mu¨hlhauser I, Berger M. Patient education—evaluation of a complex intervention. Diabetologia 2002;45:1723–33. [43] Lauver DR, Ward SE, Heidrich SM, Keller ML, Bowers BJ, Brennan PF, Kirchhoff KT, Wells TJ. Patient-centered interventions. Res Nurs Health 2002;25:246–55.