The study was conducted using the Read&Learn online platform [10] from February to June 2025. The study was preregistered in the Open Science Framework (OSF) [11].

Three groups of participants were recruited by convenience sampling:

We used the online Read&Learn tool [10] to evaluate how participants interacted with the SoF tables while answering questions about the evidence on a health intervention. The participants received the link for the survey via email and a unique single-use username and password.

The intervention involved blurring specific table cells within a digital platform, requiring participants to click to reveal information. There were no time restrictions.

The participants were presented with 2 separate screens, one containing a question and a box for their answer, and the other displaying a SoF table with blurred cells (Multimedia Appendix 1). The first SoF table column with outcomes and the headings of other columns remained visible at all times. Clicking on a cell in the SoF table revealed its content, and an opened cell became blurred when another was clicked. Moreover, 2 authors (NV and NB) independently decided on the target cells for each question, without disagreements. After answering the questions, the system presented the next question and the same SoF table, with blurred cells.

We presented 4 different SoF tables (Multimedia Appendix 1):

First, SoF1 [12] served as a practice task to introduce participants to the structure and interaction format. It contained 2 outcomes and 2 corresponding questions (Q1 and Q2), designed to be simple and intuitive. Each question had its own single target table cell.

Second, SoF2 [12] included 3 outcomes and 3 open-ended questions (Q1-Q3), each with a clearly defined target cell. All answers could be directly retrieved from the table; no additional steps were required.

Third, SoF3 [13] represented a moderate level of complexity, with 7 outcomes and 7 questions (Q1-Q7): 2 open-ended questions and 3 multiple-choice questions followed by 2 reflective questions (“Now, review the comment provided in the table and compare it to your answer to evaluate whether your perspective aligns with the comment. Would you reconsider your response?,” “If yes, explain why.”). Moreover, 3 questions had 2 target cells each, while 2 questions relied on a single target cell. This table also required the interpretation of information across 7 target cells for reflective questions.

Fourth, SoF4 [14] was the most complex, including 4 outcomes and 8 questions (Q1-Q8; 2 open-ended questions, 4 multiple-choice questions, and 2 reflective comment questions, the same as for SoF3). There were 10 target cells in total: 2 questions relied on a single cell, and 4 questions required participants to identify and interpret information from 2 different cells. Questions in this table required a higher level of cognitive effort, as participants had to actively search, combine, and interpret data from multiple table cells to be able to give a correct response.

Face validity of the experimental set-up was tested by 5 researchers from the Center of Evidence-based Medicine. Minor changes to the wording of the questions were made.

The primary outcome measure was the number of correct answers to the knowledge questions about the information in each SoF table (Multimedia Appendix 1).

Secondary outcome measures related to their search behavior: (1) total number of table cells visited during search, (2) number of target or nontarget table cells visited, (3) number of times reading the question statement, and (4) number of times searching in text for an answer. Time variables (in seconds) were (1) total time spent on the table, from beginning to the end; (2) total time for initial reading; (3) time spent on quiz questions; (4) total time reading cells during search; (5) total time reading target or nontarget segments during search; and (6) total time reading questions.

Heatmaps were created for the average times (in seconds) spent reading SoF table cells and the number of clicks (readings) for each SoF table cell.

The participants also completed group-specific demographic questionnaires tailored to their background. All participants self-reported their gender. For the purposes of analysis, responses were categorized as female or male.

We hypothesized that the expected difference between the groups would be 4 points on a scale from 0 to 16 for the number of correctly answered questions, with an SD of 5. Based on those parameters, with 80% power and a 5% α error rate, we used an online sample size calculator (Select Statistical Services sample size calculator) to calculate the sample size at 25 participants per group.

Participants, those administering the interventions, and those assessing the outcomes were aware of the study condition assignments, reflecting the quasi-experimental design. Data analysis was performed in a blinded fashion.

Each participant was a unit of analysis. Only the complete responses were included in the analysis. We assessed data normality using the Shapiro-Wilk test and presented the data as medians with IQR. We used the Kruskal-Wallis test and the Conover post hoc test to test the differences between the 3 groups. The significance level was set at P<.05. We performed the statistical analyses using MedCalc (v 23.2.1; MedCalc Software Ltd).

The results of the study are reported according to the Transparent Reporting of Evaluations with Nonrandomized Designs (TREND) reporting guidelines [15].

The study was performed in accordance with relevant regulations and guidance on social science research involving human participants. The Ethics Committee of the University of Split School of Medicine approved the study on October 24, 2024 (Document Class: 029-01/24-02/0001, registration 2181-198-03-04-24-0101). Only the lead researcher (NV) had access to participants’ personal information (name and email) during the conduct of the study for the purpose of correspondence. All others, including other researchers, only received data once it had been anonymized. The researcher retrieving the data from the Read&Learn platform had no access to participants’ identities. The output from the platform was fully anonymous, and the platform does not link the results with the entry codes.

Participants did not receive financial compensation or other incentives for participation in the study.

This study investigated how experts in evidence synthesis, clinicians, and health care users cognitively interacted with SoF tables of increasing complexity while answering intervention-related questions. Overall, we found that simpler SoF tables with clearly identifiable target information were interpreted successfully by most participants regardless of expertise. However, as the complexity of the tables and the cognitive demand of the tasks increased, performance declined across all groups. Experts demonstrated more targeted search behavior and more efficient navigation patterns, but even they experienced difficulties with tasks requiring interpretation and synthesis across multiple table elements. These findings align with our study objective of examining how different user groups search for, process, and interpret information presented in SoF tables.

We presented SoF tables with increasing complexity and increasing difficulty of knowledge questions, which increased the cognitive load presented to the study participants. Cognitive load, defined as the mental effort required to process information within working memory [16], is relevant in the context of our study, as interpreting SoF tables involves complex information that can be difficult to process, especially for health care users. Cognitive load is influenced by the complexity of the task (intrinsic cognitive load), the way information is presented (extraneous cognitive load), and the effort involved in building new knowledge structures (germane cognitive load) [16].

Intrinsic load is the inherent level of difficulty associated with learning the material or task itself [17]. All SoF tables contained statistical data, effect sizes, CIs, GRADE certainty ratings, and outcome comparisons, whose understanding required some level of statistical and methodological knowledge. The intrinsic load of SoF3 and SoF4 was high enough to challenge all groups, which likely leveled the performance across experts and nonexperts. Consequently, experts showed advantages in only 4-5 out of the 13 parameters, suggesting that previous knowledge could mitigate but not fully overcome the cognitive demands imposed by tasks with high element interactivity.

Extraneous load mainly comes from how the information is presented and the instructional design, which can raise cognitive load without improving learning [17]. Poorly designed tables (eg, unclear headings, jargon, dense text, or inconsistent formatting) increase extraneous load [17]. The varying number of cells, outcomes, and proportion of target to nontarget information in our study likely added to extraneous load, as participants had to search through irrelevant content. In addition, the Read&Learn platform required participants to reveal 1 table cell at a time, while previously opened cells became blurred again. This likely increased working memory demands because participants had to retain previously viewed information while comparing outcomes and synthesizing answers across multiple table cells. Such conditions differ from real-world reading of SoF tables, where multiple data points remain simultaneously visible and can be rapidly compared through eye movement. This may explain why performance differences between groups were limited, since higher extraneous demands made the tables more difficult to navigate, regardless of the expertise.

Our findings also suggest that comprehension of SoF tables should not be viewed as a single construct, but rather as a combination of multiple cognitive processes, including information retrieval, interpretation, comparison, recall, and synthesis. Participants generally performed well on simpler tasks requiring identification of a single target value, but had substantially more difficulty with questions requiring comparison of multiple outcomes or integration of statistical and textual information. This finding is consistent with previous literature [4,18] showing that different evidence presentation formats may support some cognitive tasks better than others.

Recent research in digital information processing and cognitive load has similarly shown that increasing informational complexity and element interactivity can negatively affect comprehension and decision-making performance, particularly in cognitively demanding environments [19].

Germane load refers to the mental effort learners invest in processing and working with the material in ways that support understanding and promote learning [20]. Giving supportive information before practice can significantly reduce unnecessary searching that does not contribute to learning [21]. In our study, the first SoF table served the purpose of familiarizing participants with the task format rather than testing performance differences. By allowing participants to focus on understanding how the tables functioned and how to navigate them, this table primarily served to prepare them for the subsequent, more complex tasks. The third and fourth tables included more complex questions, requiring participants to draw their own conclusions rather than merely locate information. This design element allowed us to observe how increased cognitive demand influenced problem-solving and interpretation across different user groups. Taken together, our findings suggest that many participants, including experts, struggle when required to move beyond basic understanding and application, especially when confronted with tasks outside their narrow domain of expertise. Experts nevertheless demonstrated more focused navigation patterns, including fewer visits to nontarget cells and shorter time spent searching irrelevant information. However, these differences likely reflect not only familiarity with SoF tables but also broader evidence-literacy skills acquired through methodological training and experience in evidence synthesis. Therefore, the observed differences should not be interpreted solely as evidence of superior usability of the SoF format itself, but rather as reflecting an interaction between table design and user expertise. These findings suggest that nonexpert users may require additional guidance, training, or alternative evidence presentation formats to efficiently identify and interpret the most relevant information within complex evidence summaries.

Although this study focuses on SoF tables, similar challenges have been observed in other forms of health communication. For example, Sander et al [22] found that design and presentation features of hospital report cards influenced users’ comprehension, and van der Mee et al [23] reported that the way laboratory results are displayed affects patient understanding. These parallels suggest that improving the clarity and usability of evidence presentation is a broader concern in health care communication, with SoF tables representing one particularly high-stakes context where comprehension directly informs decisions.

Although all participants used the same digital tool (Read&Learn), none were familiar with it beforehand, which could have impacted their interaction with the task. The Read&Learn platform had a double-screen setup, where the participants had to switch between 2 screens. This required them to remember the question they read while searching for the answer in the table, increasing cognitive demands. Also, because participants solved the experiment online without supervision, we cannot tell if they focused entirely on the tasks while solving a question, or if they experienced outside distractions that affected their reaction.

Another important limitation is that the selective cell-uncovering methodology does not fully replicate real-world interaction with SoF tables. Requiring users to click on cells to reveal information may have restricted natural search behavior and increased task difficulty compared with fully visible tables. This approach likely emphasized memory and sequential processing more strongly than ordinary reading conditions, particularly for questions requiring comparison of multiple values across outcomes. Therefore, the findings should be interpreted as reflecting cognitive interaction patterns under controlled experimental conditions rather than direct simulation of real-world SoF table use.

Differences in participants’ experience with technology and data comprehension may have also influenced performance across groups. Additionally, the wording of questions may have influenced the overall participants’ search strategies and patterns of opening cells within the tables, potentially affecting how information was accessed and processed.

The study may also have been underpowered to detect smaller intergroup differences for some secondary behavioral outcomes, particularly given the variability in time-based interaction measures. Therefore, nonsignificant findings should be interpreted cautiously.

Because our expert groups included both international and Croatian professionals, the results for experts are likely applicable to a wider population. Other groups were mainly local, and findings for those participants may not fully apply to people in different regions or settings. Furthermore, the health care user sample was relatively highly educated and therefore may not represent the broader population of health care users. Participation in the study required computer use and interaction with an online platform, which limited recruitment to individuals with sufficient computer literacy and likely introduced selection bias. Additionally, all study materials and tasks were presented in English, requiring participants to have adequate English-language proficiency. Despite these characteristics, health care users still experienced substantial difficulties with more complex tasks, suggesting that interpretation challenges may be even greater in more representative general populations.

Our findings suggest that the increasing complexity of SoF tables substantially affects users’ ability to efficiently identify, interpret, and synthesize evidence, even among experts familiar with evidence synthesis methods. While experts demonstrated more focused and efficient search strategies, expertise alone was insufficient to fully overcome the cognitive demands imposed by complex evidence summaries.

Despite being designed for clarity, SoF tables remain challenging even for experts who use them regularly, highlighting the cognitive demands of interpreting complex evidence. SoF tables should prioritize a small set of critical outcomes and present absolute alongside relative effects with certainty ratings. More interactive formats may help users find information more easily and navigate complex tables efficiently. Future research should explore user-centered and adaptive evidence presentation formats that reduce unnecessary cognitive burden while preserving transparency and methodological rigor. This is particularly important for patients and the general public, allowing nonexperts to understand the implications of different interventions, compare options, and make informed decisions based on reliable evidence.