Renal Clinical Study Participants Support Data Sharing and Use of Artificial Intelligence

Abstract

Introduction

Kidney diseases are a global health concern. Sharing data from renal clinical studies and the use of artificial intelligence (AI) can advance research in these pathologies. Understanding participant attitudes toward these practices is essential for ethical and effective implementation. This study explored their perspectives using a structured survey, based on the hypothesis that participants held positive views of data sharing and AI use.

Methods

The survey was distributed to European clinical centers. It included 42 questions assessing attitudes toward data sharing, AI use, and collecting explanatory variables. Data were analyzed using descriptive statistics, Cronbach’s alpha for internal consistency, statistical tests to evaluate the effect of selected variables, regression to identify predictors of attitudes, and principal component analysis (PCA) to explore underlying factors.

Results

Participants expressed positive views about data sharing (mean score: 0.52 ± 0.24) and AI use (0.33 ± 0.24) on a normalized scale from −1 (negative) to 1 (positive). Internal consistency was high. Regression analysis identified institutional trust and family income as predictors of attitudes toward data sharing. Health status, institutional trust, and AI knowledge were predictors of attitudes toward AI, indicating that those with better health, more institutional trust, and higher AI knowledge had a more favorable view. PCA revealed 2 distinct dimensions for data sharing (benefits and concerns) and 1 for AI (concerns).

Conclusion

Renal clinical study participants generally support data sharing and AI use, although views on AI are less favorable in patients than healthy volunteers. Institutional trust shapes attitudes in both areas, highlighting the relation between both domains.

Graphical abstract

Kidney disease is a global health challenge and is currently the third fastest-growing cause of death worldwide.^1,2 Addressing this burden requires clinical studies to evaluate potential diagnostic and treatment strategies. These studies, involving human participants, collect large amounts of valuable data, not only medical datasets and individual participant information but also study protocols, to fully understand disease progression and evaluate treatment efficacy.³ In this context, data sharing and AI practices have the potential to advance scientific research and improve outcomes.^4,5

Data sharing involves the removal of personal identifiers from clinical data, followed by the controlled release of this deidentified information to individuals or organizations outside the original research team. Recipients may include external researchers, academic institutions, or companies engaged in medical product development. Data sharing in clinical research offers many benefits. It promotes reproducibility of scientific findings by enabling independent analyses across different research centers, and it fosters collaboration by allowing data from multiple studies to be combined, facilitating meta-analysis and complex investigations, such as those needed to understand rare diseases. In addition, data sharing supports reevaluation of published studies and the exploration of new hypotheses, preventing unnecessary repetition.⁶ However, a review from 2021 highlights the ethical and economic dimensions of data sharing. Although the utility of the data is increased, it requires specific measures to protect the privacy of the participants.⁶ Their perspective is crucial in understanding the perceived benefits and barriers to sharing data. A previous work investigated data sharing perspectives at medical centers in the United States.⁵ However, their conclusions might not necessarily reflect the perspectives of renal clinical study participants in Europe because these views are context-dependent and can vary significantly across geographic regions^7,8 and they could be shaped by the different legal frameworks for personal data protection in the United States and Europe. Thus, a study specifically focused on European individuals who participate in renal clinical studies, such as those with kidney disease, is valuable to add disease-specific insight to the existing literature by capturing the particularities of this population.

AI in health care refers to the use of advanced algorithms capable of performing tasks that typically require human intelligence. In the context of clinical research, AI systems can analyze large volumes of patient data to automate and accelerate processes such as identifying patterns between treatments and outcomes, detecting lesions, or assisting in organ segmentation to improve diagnosis and treatment. These capabilities can enhance the efficiency, accuracy, and scalability of data interpretation in medical research. However, machine learning models operate probabilistically and produce predictions based on likelihoods, which can lead to the generation of imprecise or incorrect information, especially when used without adequate training or supervision, which can affect how individuals perceive and engage with the use of AI tools. Responsible use of AI in clinical research involves safeguarding sensitive data, complying with organizational policies, and ensuring the use of ethical decision-making frameworks. Understanding how participants perceive potential harms and benefits of AI is key to promoting the acceptability of the use of AI applications in clinical research. However, the perspectives of clinical study participants on AI remain underexplored.⁹

It is worth highlighting the strong link between data sharing and AI, where data sharing is essential for AI development. Researchers often rely on open-access repositories and multicenter collaborations to build and train AI models. Therefore, surveys that combine questions on both data sharing and AI within a single questionnaire can provide valuable insights into these interconnected aspects.

This study, conducted as part of the multi-center RESPECT project funded under the ERAPerMed JTC-2020 call, aimed to explore the attitudes of European renal clinical study participants toward data sharing and the use of AI in clinical research through a structured survey and applying analysis methods to facilitate interpretation. Specifically, the analysis included the following: evaluating the consistency of item responses within the survey, obtaining an index that captured participants’ overall attitude toward data sharing and AI, identifying key predictors of these attitudes through regression analysis, and examining the underlying structure of the survey items using PCA. We hypothesized that participants in European renal clinical studies would hold positive attitudes toward data sharing and AI.

Methods

This prospective study was approved by the Ethics Research Committees of the involved research centers where this approval was needed. The study enrolled individuals from 5 different countries that participated in renal clinical studies from April 2023 to May 2025, including both patients and healthy subjects.

Survey

Design

The survey started with an information section, in which participants were provided with an overview outlining the purpose and scope of the study. This section included the following: (i) the rationale for conducting the survey, emphasizing the importance of understanding participant opinions on data sharing and the use of AI in clinical data processing; (ii) a brief explanation of what constitutes a clinical study and the types of data that can be collected; (iii) definitions of data sharing and AI; (iv) a description of the survey content and estimated completion time (approximately 20 minutes); and (v) a clarification of participant rights, emphasizing the voluntary nature of participation. This information was intended to ensure that participants had a clear understanding of key concepts before responding to the survey.

Then, a 42-question survey with categorical or rating scale responses was provided (Figure 1). Many questions included multiple subquestions, referred to as items. In total, there were 114 items to assess participants’ views toward data sharing and AI. The section on data sharing, which was previously published by Mello et al.,⁵ comprised 13 questions (43 items), including a subsection that addressed how permission for sharing data should be requested in the informed consent. The AI section included 11 questions (48 items) that were formulated after consulting several sources.9, 10, 11, 12 Both sections ended with a summary question (Q26 and Q42 for data sharing and AI, respectively) assessing the general perspective of the subject on the balance between the benefits and risks of these techniques. Response options ranged from strongly negative (“The disadvantages/drawbacks far outweigh the advantages”) to strongly positive (“The advantages/benefits far outweigh the disadvantages”). In addition, the survey included 18 questions (23 items) on demographics and other explanatory variables (listed in Supplementary Table S1) potentially related to attitudes toward data sharing and AI, such as trust in people or institutions and computer skills. The complete survey text is available in Supplementary Material.

Figure 1.

Example of questions from the 3 sections of the survey. Each question includes different response options. (a) Data sharing section. (b) AI section. (c) Section on demographic and explanatory variables. AI, artificial intelligence.

Distribution

The survey, originally written in English, was translated into local languages using a back-translation method, to ensure conceptual equivalence across languages.¹³ Surveys were distributed to participants enrolled in renal clinical studies taking place in the involved research centers, in the participant local language (except in the German center, where the English version was used). By completing and returning the survey, participants provided informed consent to join the study. All responses were anonymous, and no personal information was collected. Surveys were initially distributed in person in paper format. Once completed, the responses were entered online in Google forms for further processing and analysis.

Data Preprocessing

Rating scale responses were coded numerically in ascending order, with lower values indicating negative opinions and higher values indicating positive opinions. For instance, in the example question shown in Figure 1a, the 5-point scale response options "A great deal," "A lot," "A moderate amount," "A little," and "Not at all" were coded as 5 to 1, respectively, with 1 indicating the most negative and 5 the most positive opinion. Nonordinal variables were coded arbitrarily.

Variables used to compute the subjects’ overall opinion on data sharing and AI, as well as ordinal explanatory variables used as predictors in the regression, were normalized to a scale from −1 to 1, where 0 represented the neutral point (neither positive nor negative views).

Statistical Analysis

Responses were analyzed using RStudio (version 2022.02.3; Build 492, Boston, MA).

Descriptive Statistics

Counts and percentages of the descriptive variables (including explanatory variables, responses to permission for sharing data questions and summary questions) were computed. Missing responses (including “I don’t know” answers) in data sharing and AI questions were examined for each item.

Survey Internal Consistency

Cronbach’s alpha¹⁴ was calculated separately for data sharing (22 items) and AI (47 items) sections, including questions with 5 response options.

Subjects’ Overall Opinion Index on Data Sharing and AI

To derive metrics representing the individual overall opinion on data sharing and the use of AI, response values were averaged. Responses to 35 of 43 items for data sharing (excluding summary question, questions related to permissions for data sharing, and questions 13 and 15) and 47 of 48 items for AI (excluding the summary question) were used. The Spearman correlation between this overall opinion index and participant responses to the summary question (Q26 and Q42 in the survey for data sharing and AI, respectively) was computed. The Spearman correlation between the overall opinion indexes on data sharing and AI was computed to investigate correlations between both dimensions of the survey.

Differences in the overall opinion indexes across the subjects’ demographics (gender - categorized as male, female or other and age), country of residence and participant clinical category were evaluated separately for data sharing and AI, using Kruskal Wallis tests, followed by pairwise comparisons with Wilcoxon sum rank tests and Bonferroni correction. Differences were considered statistically significant when P ≤ 0.05.

Regression Analysis

Regression analysis was performed to identify key predictors of individual data sharing and AI views. A 2-step regression analysis was employed with the subjects’ overall opinion metric defined above as the dependent variable. First, univariate Gaussian regression was conducted with each explanatory variable as independent variable, followed by multivariate regression. Explanatory variables that yielded a regression P-value ≤ 0.10 in the univariate analysis were considered as potential predictors. Pairwise Spearman correlations between ordinal explanatory variables were examined (Supplementary Figure S1). For any pair of significant variables after the univariate analysis with a correlation coefficient > 0.4, only the variable with the lower P-value in the univariate analysis was retained for inclusion in the multivariate Gaussian regression, to avoid the inclusion of correlated variables. Final predictors were considered significant at P ≤ 0.05.

PCA

PCA was performed to examine the underlying factor structure of data sharing and AI questionnaires. The same response items used to compute the subjects’ overall opinion metric were used. Incomplete response data were handled using pairwise deletion¹⁵ that computed the correlations per item pair, including only respondents who provided scores on both items; subsequently PCA was computed by carrying out an eigenvalue decomposition on the correlation matrix. To assess the suitability of the data for PCA, the Kaiser-Meyer-Olkin (“psych” R package) measure of sampling adequacy and Bartlett’s test of sphericity (“stats” R package) were conducted.

Results

A total of 204 research participants responded to the survey distributed in clinical centers in 5 different countries: 56 in Italy, 17 in Denmark, 8 in Germany, 44 in Spain, and 79 in the UK. Detailed outcomes of the recruitment process in each country are included in Supplementary Material.

The average rate of missing responses across questions was 2.43% (min: 0%, max: 14.21%) in the data sharing section questions and 12.46% (min: 3.92%, max: 35.78%) in the AI questions.

Participant Demographic and Clinical Profile

In Table 1, we show participant responses to demographic and clinical questions. Participants were primarily motivated to participate in the clinical study by altruism (67.2%) or personal health benefit (22.5%). Most respondents (76.9%) rated their health positively (61–100 points on a 0–100 scale, where 100 represents the best possible state of health, while 0 represents the worst possible state of health) and had previous experience participating in clinical studies as patients with the disease under study (57.4%). Chronic kidney disease was the most common condition (84.5%), (within this, 9% had diabetic kidney disease, 7.4% had hypertensive or ischemic nephropathy, and 2.5% had polycystic kidney disease). Most participants had a positive experience (62.3% "very positive" and 18.1% "somewhat positive"). Educational levels were high (43.1% with a university degree, 47.1% with secondary education). Participants were predominantly Caucasian (91.2%), with balanced gender and age distribution. Regarding family income, most participants (35.3%) reported earning between 25,000€ and 54,999€ per annum. Finally, only 16.2% of participants had experienced personal data theft. Detailed information on these variables is presented in the Supplementary Table S1. Responses to the questions related to permission to share data in the informed consent are presented in Supplementary Table S2.

Table 1.

Participant demographic and clinical profile

Explanatory variable	Category	Count (n)	%
Participant clinical category
	Person with studied disease	117	57.4
	Healthy person at risk	10	4.9
	Healthy volunteer	45	22.1
	NA	32	15.7
Reason to participate
	Health benefit	46	22.5
	Help others	137	67.2
	Other	10	4.9
	NA	11	5.4
Health status
	0–20	1	0.5
	21–40	7	3.4
	41–60	37	18.1
	61–80	110	53.9
	81–100	47	23.0
	NA	2	1.0
Age, yr
	< 18	0	0.0
	18–40	49	24.0
	41–60	73	35.8
	61–75	64	31.4
	≥ 76	17	8.3
	NA	1	0.5
Gender
	Male	116	56.9
	Female	87	42.6
	NA	1	0.5
Clinical condition of participant patients
	Chronic kidney disease	79	64.7
	Diabetic kidney disease	11	8.8
	Renal transplantation	3	2.5
	Acute kidney injury	1	1.0
	Polycystic kidney disease	3	2.5
	Other	6	4.9
	Heart condition	2	1.5
	High blood pressure	9	7.4
	Other genetic kidney disease	1	1.0
	NA	7	5.9

NA, no answer.

Clinical condition includes only subjects who participated in the clinical study as people with the disease under study. It may be possible that a participant has selected > 1 category. All explanatory variables included in the study are presented in Supplementary Table S1. The total number of participants was 204.

Other Explanatory Variables

Participants reported varying levels of trust in people and institutions. The majority (64.7%) trusted others “most of the time,” whereas 19.6% said “about half the time,” and 0.5% “never.” Universities (74.5%) and doctors (79.4%) received the highest institutional trust (indicated by “a lot” or “a great deal”). In contrast, trust in health insurance companies was lower, with most participants (39.7%) expressing “a little” trust. Trust in drug companies (55.9%) and government agencies (39.2%) was “moderate.” Nonprofit organizations (19.7% indicating “a lot”) and patient associations (17.2% indicating “a moderate amount”) were generally trusted.

Most respondents (87.7%) owned a phone with internet and 34.8% reported 2 to 4 hours of daily screen time. Computer skills were rated as “neither good nor bad” by 34.8%, “good” by 27.9%, and “excellent” by 11.8%. AI familiarity was limited (36.8% knew “very little,” 31.4% “a little,” and only 8.3% “a lot”). AI technical knowledge was mostly “bad” (38.2%) or “regular” (21.6%), with only 13.2% rating it “good” and 2.9% “excellent.”

Perceived benefits and concerns

The consequence of data sharing that was most concerning to participants (Q13, Table 2) was the fear that personal information might be stolen, and its impact on family, selected by 8.8% of respondents. On average, across the negative consequences (Q12, Table 3), 8% were “very concerned,” whereas 21% were “somewhat concerned.” Conversely, the most important benefit (Q15, Table 4) was accelerating scientific discovery using existing data (26.5%). Finally, in the summary question on data sharing (Q26, Supplementary Table S3), nearly two-thirds of the participants (65.6%) responded that the advantages outweighed the disadvantages (to a great, moderate, or slight extent). Another 10.6% considered them equal, whereas 14.7% believed the disadvantages outweighed the benefits (to a great, moderate, or slight extent). For data sharing (Q14, Figure 2a), 51.5% of respondents selected “a great deal” for the impact on understanding rare diseases.

Table 2.

Most concerning consequences of data sharing

Most concerning consequence of data sharing (Q13)	N	Percentage
NA	67	32.8
i. My information might be stolen, and this could affect my family.	18	8.8
g. It could be harder to get people to agree to be in clinical studies if they know their data will be shared.	15	7.4
l. Scientists and companies might have less incentive to invest time and money in doing clinical trials.	15	7.4
c. It could be embarrassing if the information was somehow related to me.	14	6.9
j. Companies might use the information for marketing purposes instead of scientific purposes.	13	6.4
e. My information might be used in scientific projects that I wouldn't approve.	12	5.9
d. People could use the data to do low quality science.	11	5.4
h. My information might be stolen, and this could affect me.	10	4.9
f. Some people or company could make plenty of money developing products with my information.	9	4.4
k. Scientists and companies might have less incentive to invest time and money in doing clinical trials.	9	4.4
a. Someone who is good with computers could identify me.	5	2.5
m. My information might be stolen, and this could increase the cost of my medical insurance.	5	2.5
b. I could be discriminated if the information was somehow related to me.	1	0.5

NA, no answer.

Table 3.

Detailed results: list of data sharing concerns

LIST OF DATA SHARING CONCERNS (Q12)
Item	Very concerned	Somewhat concerned	Not very concerned	Not at all concerned	NA
Q12a: Someone who is good with computers could identify me	4%	19%	42%	35%	0%
Q12b: I could be discriminated if the information was somehow related to me.	3%	10%	39%	46%	2%
Q12c: It could be embarrassing if the information was somehow related to me	3%	7%	32%	56%	2%
Q12d: People could use the data to do low quality science	7%	24%	40%	30%	0%
Q12e: My information might be used in scientific projects that I wouldn’t approve	7%	25%	39%	29%	0%
Q12f: Some people or company could make plenty of money developing products with my information	8%	21%	39%	31%	1%
Q12g: It could be harder to get people to agree to be in clinical studies if they know their data will be shared	9%	35%	32%	23%	0%
Q12h: My information might be stolen and affect myself	10%	17%	41%	31%	0%
Q12i: My information might be stolen and affect my family	13%	17%	36%	33%	0%
Q12j: Companies might use the information for marketing purposes instead of scientific purposes	16%	31%	34%	20%	0%
Q12k: Scientists and companies might have less incentive to invest time and money in doing clinical trials.	8%	27%	37%	27%	0%
Q12l: Scientists or companies could unfairly “free ride” on the work of others.	13%	25%	39%	22%	0%
Q12m: My information might be stolen, and this could lead to an increase in my health insurance premiums	8%	15%	36%	39%	2%
AVERAGE ACROSS ITEMS	8%	21%	37%	33%	1%

NA, no answer.

Table 4.

Most important benefits of data sharing

Most important benefit of data sharing (Q15)	n	Percentage
a. Help get answers to scientific questions faster by using information that others have already collected	54	26.5
NA	42	20.6
d. Help patients and patient groups learn more about the health problems that affect them	29	14.2
f. Support learning studies about diseases that only a small number of people have (combining data from many clinical studies)	28	13.7
i. Make sure that people's participation in clinical trials results in the greatest possible scientific benefit	26	12.7
e. Help scientists verify the accuracy of research results announced by other scientists or companies (by redoing the analyses)	8	3.9
g. Dissuade scientists and companies from hiding or distorting the results of their clinical studies (making it possible for others to check)	8	3.9
b. Help ensure that research money is spent as wisely as possible	7	3.4
c. Reduce the cost of developing new medical products	1	0.5
h. Help lawyers prove their case in lawsuits alleging that medical products are unsafe	1	0.5

NA, no answer.

Figure 2.

(a) Impact of data sharing. 51.5% of respondents selected "a great deal" for the impact of understanding rare diseases. (b) Trust in artificial intelligence. Only 2.5% of respondents answered “not at all” regarding the use of artificial intelligence algorithms in health care to analyze clinical data, such as magnetic resonance imaging. (c) Impact of artificial intelligence. Only 4.4% expressed disagreement with using artificial intelligence for processing clinical data, such as magnetic resonance images. NA, no answer.

For AI (Q34, Figure 2b), only 2.5% reported no trust (“not at all”) in its use for clinical data analysis. Most supported AI in data processing (Q33, Figure 2c), with only 4.4% indicating disagreement. However, participants were concerned with the possibility of AI devices making inadequate medical decisions (Q35, 29% rated this concern as high or very high) and leading to medical errors (28.9%). They were also worried about the use of AI tools facilitating collection of personal information (Q37, 34.8%) and the privacy of this information (31.8%). In the summary question (Q42, Supplementary Table S3), 30.9% believed AI benefits far outweighed the drawbacks, 19.1% felt that the benefits moderately outweighed them, and only 4.4% believed drawbacks far outweighed the benefits.

Survey Internal Consistency

Internal consistency was excellent with Cronbach's alpha value of 0.91 (95% Feldt confidence boundaries: 0.89–0.93] for data sharing and 0.93 (95% Feldt confidence boundaries: 0.91–0.94) for AI.

Overall Opinion on Data Sharing and AI

In Figure 3a, we show that overall opinions on data sharing and AI were positive, with average ± SD scores of 0.52 ± 0.24 for data sharing and 0.33 ± 0.24 for AI. No significant effects of gender, age, country of residence, or clinical category were found on the subjects’ overall opinion index on data sharing (Table 5). However, clinical category had a significant effect on the subjects’ overall opinion index on AI (Table 6), which had lower values for patients than healthy volunteers (patients: 0.29 ± 0.23; healthy volunteers: 0.40 ± 0.22). Significant differences were found on this index between participants from Spain and UK (Spain: 0.40 ± 0.23; UK: 0.27 ± 0.26).

Figure 3.

(a) Subjects’ overall opinion index on data sharing and AI. Histograms are presented on a scale between −1 and 1, with 0 representing the neutral point. (b) Correlations between the subjects’ overall indexes of data sharing and AI. (c) Association between data sharing overall opinion index and response to the summary question. (d) Association between AI overall opinion index and response to the summary question. AI, artificial intelligence; NA, no answer.

Table 5.

Effect of selected variables on the subjects’ overall opinion index: data sharing

Variable	Kruskal-Wallis test	Pairwise comparisonsa
Gender		Male vs. female; P = 0.637
Age	χ2 = 6.26, df = 3, P = 0.099
Clinical category	χ2 = 0.71, df = 2, P = 0.699
Country of residence	χ2 = 2.35, df = 4, P = 0.672

^aPairwise comparisons were performed using Wilcoxon rank sum test with Bonferroni correction.

Table 6.

Effect of selected variables on the subjects’ overall opinion index: AI

Variable	Kruskal-Wallis test	Pairwise comparisonsa
Gender		Male vs. female; P = 0.086
Age, yr	χ2 = 1.38, df = 3, P = 0.709
Clinical category	χ2 = 9.11, df = 2, P = 0.010	Person with studied disease vs. healthy volunteer; P = 0.03 Person with studied disease vs. healthy person at risk; P = 0.16 Healthy person at risk vs. healthy volunteer; P = 1
Country of residence	χ2 = 9.82, df = 4, P = 0.043	Spain vs. UK; P = 0.05b

AI, artificial intelligence; df, degrees of freedom.

Significant P-values are ≤ 0.05.

^aPairwise comparisons were performed using Wilcoxon rank sum test with Bonferroni correction.

^bOwing to the large number of comparisons, only significant results are reported.

In Figure 3b, we show the correlation between the subjects’ overall opinion indexes on data sharing and AI (Spearman ρ = 0.425, P < 0.0001). In Figure 3c and d, we show the correlations between the overall opinion index and the response to the summary question for both data sharing and AI, respectively. For both data sharing and AI, there was a trend; as participants perceived more the advantages than disadvantages of the technique, the subjects’ overall opinion index increased. Spearman’s rank correlation coefficients indicated a statistically significant positive correlation for data sharing (ρ = 0.344, P < 0.001) and AI (ρ = 0.482, P < 0.001).

Regression Analyses

After univariate analysis, the following explanatory variables were considered as potential predictors of the overall opinion index for data sharing (P < 0.1): reason to participate, experience in the clinical study, education, family income, people trust and the institutional trust average, AI knowledge, and AI technical knowledge. Items related to institutional trust were positively correlated (Spearman’s ρ = 0.43–0.51, Supplementary Figure S1), and their mean value was computed before their inclusion in the univariate regression analysis. People trust and institutional trust average were also correlated (Spearman’s ρ > 0.4); thus, only the variable with the lower P-value (institutional trust) was retained. AI knowledge and AI technical knowledge variables were correlated, and only AI technical knowledge was retained. In multivariate analysis, institutional trust (P < 0.001) and family income (P < 0.0192) were significant predictors. Higher institutional trust and income were associated with more willingness to share data.

For AI, univariate analysis identified reason to participate, current health status, participant clinical category (i.e., person with the studied disease, person at risk of disease, or healthy volunteer), experience in the clinical study, education, family income, people trust, institutional trust, AI knowledge, and AI technical knowledge as potential predictors of the overall opinion index. Among these, health status and clinical category were positively correlated (Spearman’s ρ > 0.4), as were people trust with institutional trust average, and AI knowledge with AI technical knowledge. In each correlated pair, only the variables with the lower P-value in the univariate analysis (health status, institutional trust average, and AI technical knowledge, respectively) were retained. Multivariate analysis showed that health status (P = 0.013), institutional trust average (P = 0.0008), and AI technical knowledge (P < 0.049) were significant predictors, indicating that those with better health status, more institutional trust, and higher AI technical knowledge had a more favorable view of AI. More detailed information of the regression analysis is presented in the Supplementary Material.

PCA

The Kaiser-Meyer-Olkin measure (data sharing: Kaiser-Meyer-Olkin = 0.87; AI: Kaiser-Meyer-Olkin = 0.7) and Bartlett’s test of sphericity (data sharing: Χ²(34) = 1174.5, P < 0.001; AI: X²(46) = 199.7, P < 0.001) indicated that the sample size, and the data were adequate for conducting PCA.

Data sharing PCA revealed 2 components that explained a total of 45.9% of the variance (Figure 4a). Items with high loadings on component 1 (> 0.30) are related to the benefits of data sharing and are primarily associated with positive perceptions. The second component was linked with concerns about data sharing, with high loading (> 0.30) for items reflecting potential risks or apprehensions (Figure 4b).

Figure 4.

(a) PCA variance explained by all principal components for data sharing and artificial intelligence questions. (b) Loadings for the first 2 principal components. PCA, principal component analysis.

AI PCA two components explained 31.4% of the variance (Figure 4a). The first component was primarily associated with perceived benefits of AI, whereas the second captured concerns with AI risks. However, the presence of high loading across additional items (Figure 4b) reflects that the survey was designed to assess multidimensional aspects of AI, that is, not only perceptions of benefits and risks but also opinions on other issues such as efficiency.

Discussion

This study explored perceptions of data sharing and AI among participants in renal clinical studies conducted 5 countries. We hypothesized that participants in renal clinical studies would show generally positive attitudes toward data sharing and AI. The findings confirmed this hypothesis, revealing institutional trust (for both data sharing and AI), family income (for data sharing), health status (for AI), and technical knowledge of AI were significant predictors of positive attitudes.

Participant Profile

Majority of respondents (62.3%) reported highly positive experiences as participants, consistent with findings reported by Mello et al.,⁵ which can be related to the fact that participants in clinical studies are intrinsically motivated to contribute to scientific research.

Data Sharing Perceptions and Predictors

Participants in this study generally expressed positive perceptions toward data sharing. As reflected in the summary question, 65.6% of respondents indicated that the advantages outweighed the disadvantages. This proportion is somewhat less than the 82% reported by Mello et al.⁵ among American participants. In addition, 14.7% of respondents in the present study indicated that they believed that the disadvantages outweighed the benefits, compared with the 8.4% in Mello et al.’s⁵ findings. These differences may reflect regional variations in trust or cultural attitudes toward privacy, that could have evolved in the years since that study was conducted.

Across the negative consequences assessed in the survey, the average proportion of respondents who were “very concerned” was 8%, consistent with the findings of Mello et al.,⁵ and 21% of respondents in this study reported being “somewhat concerned,” slightly higher than the 17% reported by Mello et al. This suggests a moderate concern in our sample. A substantial minority expressed some concern that data sharing could make it harder to recruit participants for clinical studies (35%) or that companies might misuse the information for marketing rather than scientific purposes (31%). These concerns are in agreement with those reported by Mello et al.⁵ and Mozersky et al.¹⁶

Participant demographic characteristics and country of residence did not influence the subjects’ overall opinion index on data sharing. There were no differences either across the 3 clinical categories, with very similar average values for the 3 groups. This was unexpected because patients are more likely to benefit from studies performed on shared data. However, it could reflect the characteristics of the healthy volunteers who were participants in research studies, and are known to be motivated by their interest in contributing to science and improving the health of others.¹⁷ This agrees with the selection of “help get answers to scientific questions faster” as the most important benefit of data sharing and could explain their high willingness to share their data.

The regression analysis revealed differences in data sharing views depending on the levels of institutional trust and family income. Participants with higher trust and income were more willing to share their data. This finding aligns with previous studies in genomic research,¹⁶ which highlight trust in institutions and researchers as key factors in participants’ support for data sharing. The association with income is in line with previous work suggesting that low- and -middle income populations may face additional barriers and challenges related to data sharing.¹⁸

The PCA analysis showed that the survey instrument is structured around 2 core dimensions are related to data sharing: perceived benefits (such as support for science) and concerns (such as privacy and potential misuse). These 2 dimensions have been widely mentioned in the literature as key components in shaping participants’ attitudes, often described as a privacy-utility trade-off,¹⁹ where individuals weigh potential benefits against concerns.

The positive attitudes toward data sharing expressed in this survey highlight the fact that objections to data sharing by research study participants are not a major issue. Indeed, most respondents supported giving broad, 1-time consent, which would be easy to obtain during a research study. Thus, to promote data sharing, efforts should focus on surmounting other more significant challenges, such as concerns from researchers themselves and lack of available infrastructures for sharing data.²⁰

Attitudes Toward AI in Clinical Research

Overall, respondents in this study expressed support for the use of AI in clinical research, though slightly less than their support for data sharing possibly because of limited AI technical knowledge. Similarly, the findings of Zhang and Dafoe²¹ reported a mixture of support and concerns regarding AI, with 42% of participants supporting the use of AI and 22% opposing it.²² These figures are comparable with our results, where 50% of respondents believed the benefits of AI outweigh the drawbacks, and only a small minority (8.3%) felt that the drawbacks outweighed the benefits.

Patients had a less favorable view of AI than healthy volunteers, as reflected in the results of statistical comparisons across clinical categories (Table 6) and regression analysis, which identified better health status as a significant predictor of positive opinion on AI. This result is in agreement with previous work that has investigated patient apprehensions about the use of AI in health care, revealing multiple concerns related to AI safety, increases in medical expenses, limitations of their ability to make their own health care decisions and data security.²³ Some of these apprehensions align with the worries expressed by the participants of this study. This is concerning, because analyzing large amounts of clinical data with AI holds great potential to improve health care outcomes and patients should be the major beneficiaries of these improvements.

The regression analysis identified technical knowledge of AI and institutional trust, as significant predictors of positive attitudes. This agrees with Zhang and Dafoe²¹ who reported that support for developing AI was greater among those who have experience with technology (in particular, those with computer science or programming experience). Although age and health variables were not directly correlated, we observed modest but significant correlations between health and other factors, such as education (ρ = 0.27), and screen time (ρ = 0.28). These correlations may suggest that individuals in better health tend to have more exposure to technology, which could contribute to greater comfort with AI. This highlights the need for adequate education on the possibilities of AI, providing usable knowledge adapted to patient competence and ability, to achieve the goal of integrating AI in clinical research and practice while fostering a patient-centered model of health care.²⁴

As observed in the context of data sharing, participants with higher levels of institutional trust were more likely to support AI. This aligns with the existing literature showing high trust in universities, research institutions, and defense organizations to both develop and use AI (71%–77%) and to regulate and govern it in the public interest (67%–73%).²⁵ The consistent role of institutional trust in both domains (data sharing and AI) highlights that trust in institutions may be an important factor in public acceptance of data sharing practices and AI technologies.

PCA of the AI section revealed 2 main components in the survey (benefits and concerns); however, high loading across additional items reflects the multidimensional design of the survey. This aligns with a systematic review that identified 7 thematic dimensions in AI perceptions, such as knowledge or familiarity, benefits or risks, acceptability, development, implementation, regulation, and human–AI relationships.²⁶

Limitations

Participants were residents of 5 European countries, with different languages and cultural exposures. In addition, these 5 groups were of different sizes and they were not matched by demographic or clinical characteristics. The unequal distribution of responses across countries may have introduced geographic bias and underrepresent certain populations. Nonetheless, differences were not found across countries in the subjects’ perception of data sharing, whereas the views on AI were different only between Spain and UK. This supports the generalization of the results to other populations in developed countries. In addition, the high Cronbach’s alpha values showed excellent internal consistency, supporting the adequacy of the current sample size for the analyses.

The sample was predominantly Caucasian. Future studies should aim for more balanced and diverse recruitment and could consider including subjects not actively participating in research studies. The study will continue to be distributed to increase sample size and strengthen the generalizability of the findings.

Conclusion

The survey demonstrated high internal consistency, indicating reliable measurement of participants' perceptions. Overall, participants involved in renal clinical studies expressed positive attitudes toward data sharing and AI in clinical research, which should serve as a motivating factor to lift other barriers that currently exist to data sharing. Key factors influencing these attitudes included institutional trust, family income, health status, and technical knowledge of AI. These findings underscore the importance of building institutional trust and promoting education on AI to enhance engagement, especially in patients.

Disclosure

The authors declare no conflict of interest.

Supplementary Material

Supplementary File (PDF)

Figure S1. Correlation matrix between normalized ordinal explanatory variables. Table S1. Explanatory variables included in the survey. Table S2. Opinions on consent for data sharing. Table S3. Perceived benefits and concerns of data sharing and AI. Regression analysis. Outcomes of the recruitment process. Full text of survey questionnaire.