2.2. Data Analysis

Each evaluation of one medicinal product, taken by one patient, corresponded to a particular combination of an observed measure (e.g., fully taken) for each of the eight aforementioned observational variables (e.g., result of the intake) which describe the many aspects of acceptability. In total, 2004 evaluations were collected. A multivariate analysis mined this large set of standardized evaluations to summarize the main information into an intelligible tool: the acceptability reference framework.

The key relationships between the observed measures were visualized in a low-dimensional Euclidean space: the 3D acceptability map. A multifactorial process, multiple correspondence analysis (MCA), established the relationships between the observed measures that were often selected together in the evaluations, such as a “short time” and “positive reaction” or “use food/drink” and “use divided dose”. Such measures, as well as dots representing the evaluations completed in a similar manner, converged on the acceptability map. Thus, proximity on the map reflected a similarity. The three dimensions illustrated by the map summarized those associations and dissociations that most contributed to the total variance of the dataset (inertia). The first axis is the most important dimension, resuming 20.4% of the inertia, the second axis the next most important, summarizing 13% of the inertia, while the third dimension resumed a further 9.9% of the observed variations. Thus, the map highlighted the major information in terms of medicine acceptability variations.

Afterward, a clustering process gathered the most similar evaluations—the closest on the map—into two coherent clusters. The positive observations naturally emerged in the first cluster, defining the “positively accepted” profile, while all the negative observations were over-represented in the second cluster, defining the “Negatively accepted” profile. The profiles were materialized by a green and a red area on the map.

The evaluations of tablet intake were successively partitioned into two subgroups according to the patient’s ability to swallow and the size of the tablet taken:

the older patients without swallowing disorders (SD−) and the older patients with swallowing disorders (SD+) who had taken tablets, regardless of their size;

the older patients SD+ had taken tablets smaller than a given threshold and the older patients SD+ had taken tablets which are equal and larger than the threshold. Size thresholds increasing by steps of 0.5 mm were explored from 6 mm to 10 mm.

In each case, both subgroups of interest were positioned on the map, at the barycenter of their evaluations. If a barycenter, along with the entire 90% confidence ellipsis surrounding it, belonged to the green area of the map, the subgroup could be classified as accepted (see the video abstract for an illustration of the mapping, clustering, and scoring processes). A minimum of 30 evaluations are required to obtain a reliable acceptability score.

In each case, statistical tests were used to assess the significance of the differences observed between the two subgroups of interest in terms of patients’ characteristics, products’ features and measures composing the acceptability scores. For the categorical variables, when there was a minimum expectation of 5 for 80% of the cells without any null expectation, Pearson’s chi-squared test was used; alternatively, Fisher’s exact test was used. A Student’s t-test was used for the quantitative variable, i.e., to compare the mean tablet sizes evaluated between the two subgroups of patients, those with and without reported swallowing difficulties.