Knowledge about age-related eye diseases in the general population in Germany

Survey participants were recruited from the adult German general population to assess their knowledge about cataract, glaucoma, AMD and diabetic eye disease regarding risk factors, prevention strategies, and treatments, using an online recruiting platform (SurveyEngine, Berlin, Germany). Individuals from the general public can register on this platform for participating in surveys and are compensated according to agreements between the participants and the service provider. Eligibility criteria were age ≥ 18 years, German language comprehension, residence in Germany, and internet access. No participants were recruited from other sources than the recruiting platform.

The target size of the surveyed sample was 1,000 participants, considering a population size > 10,000 individuals (adult Germans) and a ± 3% margin of error [27]. The age and sex of participants reflected publicly available data from the German national micro-census 2021 in adults [28], which was implemented in the targeting algorithm of the service provider. Our survey included data on sociodemographic characteristics (age, sex, marital status, level of education, employment, living situation) as well as items on participants’ overall and ocular health (including the EQ-5D-5L questionnaire [29]), use of the healthcare system (number of general practitioner and ophthalmologist consultations over the past 12 months), and sixteen items on knowledge about four common eye diseases (cataract, glaucoma, AMD, diabetic eye disease).

To obtain the knowledge items, a list of contents of existing knowledge assessments instruments on the mentioned ocular conditions targeted at the general population was compiled based on a review of the literature, using the literature search tools PubMed and Google Scholar. Content of existing, published instruments was reviewed by two of the authors (JHT, RPF) and a list of draft items was produced. Four items per disease category were then selected, considering knowledge domains with a clinical or public health relevance. Each item contained a statement about a risk factor, prevention strategy, treatment or public misconception, respectively. Before the instrument was used in the quantitative part of the study, three debriefings with members from the general public were performed to ensure the comprehensibility of the items and response options. We asked participants to rate all statements as true or false (forced response) from their knowledge without support from third parties and with no additional information. Data collection took place in September 2022. Ethics approval was obtained from the institutional ethics committee at the University Hospital Bonn, Germany (approval ID 255/22).

The distribution of variables was graphically inspected and described using the mean and standard deviation (SD), absolute numbers and percentages. In the first part of statistical analysis, we investigated the internal consistency and dimensionality of the 16 knowledge items, using a dichotomous Rasch model, i.e. a one-parameter item response model, under the assumption that the individual items (questions) share a common latent trait [30]. Rasch models describe the probability of a correct response to a questionnaire item as a function of test taker ability and test item properties. We then used the models to assess person reliability, person separation index, item fit, and differential item functioning (DIF) by sex, including female and male participants. DIF was evaluated using contrasts and Rasch-Welch probabilities. Contrasts ≥ 0.64 logits were interpreted as potential DIF and contrasts ≥ 1.0 were interpreted as relevant DIF [31]. Multiple correspondence analysis (MCA) based on a correspondence analysis of the Burt matrix was conducted to further assess knowledge dimensions across disease domains. MCA is a dimension reduction technique similar to principal component analysis, specifically developed for use in categorical data [32]. The term Burt matrix refers to a cross-tabulation of all response categories. The subsequent analysis focused on the identification of sociodemographic determinants of participants’ knowledge about eye diseases. For this purpose, correct answers to the statements about eye diseases were summed up to a total score (maximum score 16 points) and sub-scores per disease (maximum score 4 points). Participants’ sub-scores were compared using the Friedman test. Associations between sample characteristics and test performance were investigated, using logistic regression models with the dichotomized sum score of correct responses by median as the dependent variable. Statistical analyses were performed with SPSS Statistics, version 27 (IBM Corporation, Armonk, NY), Winsteps, version 3.92.1 (Chicago, IL) and R, version 4.2.2 (Vienna, Austria). P-values < 0.05 were considered statistically significant.

The person reliability and person separation index in a dichotomous Rasch model from the responses were 0.46 and 0.92, respectively. Mean-square values ranged between 0.84 and 1.14 in the inlier-sensitive fit statistic and between 0.80 and 1.77 in the outlier-sensitive fit statistic. The person-item map indicated that the difficulty of the 16 items was well targeted to our population, with item A4 being most difficult and item C4 being least difficult (Fig. 1). One item (A3) was indicative of DIF for sex, with a higher probability of correct replies in female than in male participants (p = 0.0001, DIF contrast 0.69). The item was retained as the contrast threshold of 1.0 was not exceeded. Since the model requirements were not met and did not improve with any revisions, we decided not to use the Rasch model for further analysis.

In MCA, the first two dimensions explained 37.6% of the variance in the dataset (Fig. 2). Applying a rescaling algorithm that fit off-diagonal submatrices yielded 52.7% of the variance explained by the first two dimensions and a comparable biplot to Fig. 2 (data not shown). Most correct responses loaded negatively on the first dimension and all items loading positively on the second dimension included correct statements while no associations with the eye condition or the field of knowledge were qualitatively identified. As a result of MCA, all 16 items were used for further analysis.

Five hundred twenty-four participants (52.0%) had nine correct responses or less (lower-knowledge group) and 484 participants (48.0%) had ten or more correct responses (higher-knowledge group). The median number of correct responses in these subgroups was 8 and 11, respectively. Based on this we dichotomized the dataset. Binary-logistic regression analysis revealed significant associations of being in the higher knowledge group with older age, retirement, reduced general health and consultations with general practitioners and ophthalmologists within the previous 12 months, while the level of education was not significantly associated (Table 3). These results remained overall consistent when looking at the best- and worst-performing participants of the sample by the highest- and lowest knowledge quartiles instead (Supplementary Table 1) and when excluding participants with self-reported cataract, glaucoma, AMD and diabetic eye disease (Supplementary Table 2).

In a single multivariable binary-logistic regression model, higher knowledge was significantly associated with both consultations with a general practitioner and an ophthalmologist over the past 12 months, when controlling for participants’ age, sex, employment status and general health (Table 4). On average, participants who had consulted a general practitioner or an ophthalmologist replied correctly to one of the statements more than participants who had not consulted these medical specialties, respectively. Analyses at an individual item level were consistent with these results (Supplementary Fig. 1).