Study design and population
This cross-sectional study is based on data from the population-based KORA (Cooperative Health Research in the Augsburg Region) Health Survey 2016 in Southern Germany (KORA GEFU 4 study). KORA has been described in detail elsewhere [13]. KORA is a regional research platform in southern Germany for population-based health surveys. The research platform aims to continue and to expand the project ‘Multinational Monitoring Trends and Determinants in Cardiovascular Disease’ (MONICA) initiated by the World Health Organization in 1984 [14,15,16]. It examines the links between health, disease, and the living conditions of the population.
In 2016, all 11,189 eligible respondents from the KORA S1–S4 studies (from 1984 to 2001) were invited to participate in the KORA GEFU 4 study (n = 9035 responses). An additional diabetes-related questionnaire was sent by post or collected by telephone in 2016 to all eligible Health Survey respondents who reported a diabetes diagnosis (n = 1130). The present study included all 837 participants (74.1%) who responded to the diabetes-related questionnaire between May 2016 and January 2017.
Measurement of diabetes-related information needs
To assess the diabetes-related information needs of individuals with Type 1 and Type 2 diabetes, we used two different sections of the Information Needs in Diabetes Questionnaire (Additional file 1, Appendix 1) [17]:
-
1
In the first section, respondents were asked to select up to three of 11 diabetes-related topics on which they currently needed more information (multiple answers). This enabled us to identify topics where participants currently had the greatest desire to obtain more information.
-
2
In the second section, for each of the 11 diabetes-related topics, respondents were asked whether they would like to have more information on each topic at the current time. Response categories were ‘yes’ and ‘no’. Thus, the participants’ information needs were measured for each topic, without being assessed as more or less important, as in the first section.
Measurement of associated characteristics
Based on the literature (Additional file 2, Appendix 2), we selected characteristics that might be associated with information needs and defined the following six thematic groups of variables:
Sociodemographic characteristics [8, 9]
We included age (in years), sex, and years of education (primary education < 11 years vs. secondary/tertiary education ≥11 years).
Diabetes-related characteristics [9, 10, 18, 19]
We included type of diabetes, coded as ‘Type 1 diabetes’, ‘Type 2 diabetes’ and ‘other diabetes type’ (e.g. gestational diabetes; not included in the LCA models), and diabetes duration as measured by the question ‘Have you been diagnosed with diabetes by a physician? In which year?’. Antihyperglycaemic medication was coded as ‘yes’ if respondents stated that they currently took oral glucose-lowering drugs or insulin. If both treatment options were answered with ‘no’, the variable was coded as ‘no’. Diabetes-related comorbidities were coded as ‘yes’ if respondents reported at least one of the following comorbidities (retinopathy, blindness, microalbuminuria, kidney failure, artificial kidney, peripheral artery occlusive disease, polyneuropathy, diabetic foot syndrome, amputation). Otherwise, no comorbidity was assumed and the characteristic coded as ‘no’.
Lifestyle-related characteristic [9, 20]
Smoking behaviour was assessed in terms of the current smoking situation, with the responses ‘yes’ (regularly or occasionally) or ‘no’ (never- or ex-smoker).
Well-being [10, 21]
Well-being was measured with the German version of the World Health Organisation-Five Well-Being Index (WHO-5) [22] and coded as ‘low well-being’ or ‘high well-being’ (cut-off score ≥ 50).
Current level of information and diabetes education [10]
The respondents’ current level of diabetes-related information was captured using the Information Needs in Diabetes Questionnaire [17]. Responses for the 11 topics (Additional file 1, Appendix 1) were given on a 4-point Likert scale from ‘not informed at all’, coded as 0, to ‘very well informed’, coded as 3. The sum of all variables ranged from 0 to 33. In addition, we included participation in a diabetes training programme (yes vs. no) as a measure of diabetes education.
Time preference [9, 23]
To measure time preference, we asked respondents to indicate their level of agreement to the statement: ‘My present well-being is more important to me than my future health status’ on a 4-point Likert scale which was then dichotomised and coded as ‘rather disagree’ or ‘rather agree’. Time preference can be regarded as present orientation as it assessed ‘whether participants preferred immediate pleasure over long-term health’ [24].
Statistical analysis
The descriptive analysis included frequencies, mean values and standard deviations. All quantitative analyses were performed in the SAS software, V.9.4 (SAS Institute Inc., Cary, NC).
Analysis of information needs using LCA: handling of missing data
In total, 708 participants (84.6%) responded to at least one topic of the Information Needs in Diabetes Questionnaire (Section 1 or Section 2, see ‘Measurement of diabetes-related information needs’). The LCA [25] was performed with the data from the second section. Four hundred eighty participants (57.3%) provided information for at least one of the 11 topics in Section 2 (a question asking whether the participant needs information on each topic, with yes/no as the possible responses), while 283 respondents (33.8%) provided information for all 11 topics.
The missing values seemed to be not at random: Frequency of answering ‘yes’ (reporting an information need) increased for all 11 questions as the total number of missing values increased (Additional file 2, Appendix 3). Therefore, it seemed that some respondents understood this section more as a checklist in which it was not necessary to select the answer ‘no’; one only needed to answer a question with ‘yes’ if information on a given topic was needed. Hence, missing values can be assumed to at least partially reflect the answer ‘no’ (‘checklist misconception effect’ [26]).
Therefore, we decided to perform a main analysis and two sensitivity analyses in which the missing values were handled differently. In the main analysis (Variant 1), we did not perform any imputation at all. We included only respondents who answered at least one of the 11 questions of Section 2 of the questionnaire (n = 480) and used the answers that were available. In the first sensitivity analysis (Variant 2), all missing values for the 11 questions of Section 2 were coded with ‘no’ and the LCA was performed with the full sample (n = 837). In the second sensitivity analysis (Variant 3), missing values were coded as ‘yes’ if the respective diabetes-related topic was selected as one of the three most important topics in the first section of the Information Needs in Diabetes Questionnaire. We assumed that participants who had already reported a need for information in the first part of the questionnaire might not want to mention it again. This led to a total of 613 participants in Variant 3.
LCA without covariates: identification of subgroups with different information needs
In line with Lanza and colleagues (2007), we calculated LCA models without covariates with one to eight classes in order to identify the optimal number of classes [25]. We chose the best model based on model fit indicators. Lower values of the Bayesian information criterion (BIC) and adjusted Bayesian information criterion (aBIC) indicated better fit. A (relative) entropy close to one indicated high separation of classes. Moreover, we assessed whether the classes were meaningful, set the minimum class prevalence to 5%, and took the ‘law of parsimony’ [27] into account.
LCA with covariates: identification of associated characteristics
After selecting the number of classes, we performed LCA with covariates to investigate the characteristics associated with the identified subgroups. LCA with the following covariates were performed: age, sex, years of education, type of diabetes, diabetes duration, antihyperglycaemic medication, diabetes-related comorbidities, current smoking behaviour, well-being, diabetes education, current level of information, and time preferences (see ‘Measurement of associated characteristics’).