This research is part of the larger study iCARE4OLD funded by the European Union in the Horizon 2020 framework [16]. The study used a cross-sectional design using mixed closed and open-ended survey questions to explore the use and perception of CDSS among health care professionals. The protocol of the study was published recently [18]. Data collection took place simultaneously in seven countries: Belgium, the Czech Republic, Finland, Italy, the Netherlands, Poland, and the United States. The study was ethically approved in the participating countries and all health care professionals gave their written informed consent for participation. The protocol describes the different phases of the study: the pre-questionnaire phase, the patient evaluation phase in the iCARE tool, and the post-questionnaire phase. In this study, data from the prequestionnaire phase were used for the analysis.

Participants were healthcare professionals involved in the care of older adults (aged 65 years or older) with CCC, either in long-term care facilities or primary care. Eligible participants were required to be qualified professionals capable of making or proposing therapeutic decisions (eg, physicians, nurses, and physiotherapists) and interpreting interRAI assessments. We used purposive sampling combined with snowball sampling of at least 20 participants per country to ensure sufficient observations for analysis. This sampling was performed by each country’s local study lead among purposively selected professionals. To ensure consistency across countries, the study leads held regular meetings reporting the recruitment and data collection procedures.

Local coordinators followed a standardized, documented procedure: site liaison and approval; targeted invitations with a translated information sheet; brief eligibility prescreening against the four inclusion criteria; scheduling of sessions; written informed consent; and administration of a uniform testing protocol (prequestionnaire, iCARE testing session, decision‑loop where applicable, postquestionnaires).

Data collection started in November 2024 and ended in January 2025. Clinicians completed an online questionnaire created for the study, consisting of background questions (socio-demographic information, as well as information about education, profession or position, years of professional experience, etc), questions related to the use of interRAI instruments, attitudes toward new technology, and both closed and open-ended questions on the use of CDSS. The English version of the questionnaire can be found in Multimedia Appendix 1. Local study leads and their teams translated the surveys into the local languages. Using both closed and open-ended questions can result in triangulation, making the findings more valid and permitting a deeper understanding of the data [19,20].

The data were validated to ensure data integrity. Validation procedures included verifying that each respondent ID had a single set of responses across all corresponding questionnaires. Timestamp checks were conducted to confirm consistency with respondents’ records. The dataset was systematically reviewed to identify and correct any erroneous values.

Quantitative data were analyzed using descriptive statistical methods, bivariate analysis, and regression models to explore associations. Pearson correlations were used for continuous data and Spearman correlations for Likert scales. A linear regression model was built for being comfortable with technology and a logistic model for the actual use of CDSS. Age was used as a continuous variable in the linear model and as dichotomous variable (age<50 or age≥50 years) in the logistic model. Only two professions were used in the models to allow for comparison (nurse and physician). The software STATA SE (version 18.5; StataCorp LLC) was used in the analysis.

All qualitative open-ended responses were first translated into English by the local teams, and then all answers were thematically analyzed by an experienced researcher and independently reviewed by another researcher to ensure consistency. Any discrepancies were discussed collaboratively until consensus was reached. In case of ambiguity or unclear answers, the researchers discussed the interpretation to reach a consensus on classification. If no reliable interpretation could be made, the response was noted as unclear and excluded from thematic coding.

We systematically clustered responses from the 3 open-ended questions. These questions were (1) use of decision support systems, (2) use of predictive decision support systems, and (3) decision support tools participants considered helpful. We classified the reported CDSS as either “knowledge based” or “non-knowledge based,” as “diagnostic” or “prognostic,” and lastly into consensus-based categories based on the thematic analysis [2,8]. For the classification into either diagnostic or prognostic, we followed previously proposed definitions [10]. Diagnostic refers to CDSS which classify the current health status or event, whereas in a predictive CDSS, the outcome is a future event. Where possible, we searched for additional information on the reported CDSS systems to complement available information. This way, the CDSS could be classified more reliably.

Ethical approvals were granted in each participating country before the start of the pilot. Italy: Comitato Etico Territoriale Lazio 3 (ID7068); the Netherlands: Amsterdam UMC Toetsingscommissie (2024.0527); Belgium: Commissie Medisch Ethiek van de Universitair Medische Ziekenhuizen KU Leuven (No. S69591); Czechia: Ethics Committee of the Faculty of Pharmacy, Univerzita Karlova, H2020-965341,); Poland: the Research Ethics Committee at the Jagiellonian University Medical College (118.0043.1.237.2024); the United States: Advarra Institutional Review Board (No. 965341); Finland: Well-being Services County of North Karelia (6970/13.00.01/2024, § 5/2024 and Kainuu Metti Järvikallio (6970/13.00.01/2024, § 23/2024); Central Uusimaa (KEUHDno-2024-3384, § 87); and Helsinki (HEL 2024-010604 t 13 02 01). All participants gave their written informed consent to participate in the study, and data were anonymized and further pseudomized to link assessments. Some countries offered financial compensation to participants as described in the protocol of the study [14].

Table 1 shows the characteristics of the 151 participants from seven participating countries. Clinicians’ mean age varied from 38.4 (SD 9.7) years in the Dutch sample to 51.6 (SD 10.3) years in the sample from the United States. Participants from Finland and the United States were predominantly female (19/20, 95%), while the proportion of women in the overall sample was 71.5% (108/151). The proportion of participants from home care or long-term care services was similar across 6 countries (47-53%), with the exception of the Czech sample, with 28% (7/25) professionals from home care and 72% (18/25) from long-term care.

Amongst participants, physicians were the most prevalent (85/151, 56.3%), followed by nurses (57/151, 37.7%). Most respondents had a master’s educational level (99/151, 65.6%), followed by 15.2% (23/151) with a bachelor’s degree and 13.9% (21/151) with a PhD. Clinical experience was most extensive in participants from the United States (24.7 years) and Belgium (21.9 years), and relatively lower in those from the Netherlands (12 years). Similar results were found for the experience with complex care for older people, with an average of 23.1 (SD 14.9) years in the sample from the United States and 10.8 (SD 7.8) years in the Dutch sample.

Regarding users’ comfort with technology, the average score was 3.9 out of 5, with the score ranging from 1 (not comfortable at all) to 5 (very comfortable). Participants from Italy (4.3) and the Czech Republic (4.4) reported higher levels of comfort, while participants from the Netherlands reported the lowest (3.3). The average overall openness to integrating new technologies was high (4.2/5), with the highest levels for professionals from the United States (4.7), Italy (4.6), and Poland (4.5).

Table 2 shows the correlations between the most important determinants of CDSS use. A positive and significant correlation was found between being open to adopting new technologies and being comfortable with using technology in the overall sample (0.579; P<.001). This was also found for health care professionals from Belgium, the Czech Republic, Finland, and Italy, with very strong correlations in the samples from the Czech Republic (0.933; P<.001) and Finland (0.943; P<.001). All other correlations were not found to be significant in the full sample. In addition, for the participants from Belgium and the Czech Republic, there was a positive association between being open to technology and the use of CDSS. In the United States’ sample, the association between the number of years of clinical experience and the use of predictive CDSS was negative and significant (–0.464; P=.04).

A total of 78 (51.3%) participants reported using any type of CDSS in their daily work, while 34 (22.4%) reported using specific predictive CDSS. The rates for CDSS use in daily work varied across countries’ samples ranging from 15% (3/20) in Italy, 35% (7/20) in Belgium, to the highest rate of 90% (18/20) in Finland. Regarding the use of CDSS for predictions, the lowest rates were found in the Polish (1/20, 5%) and Belgian (2/20, 10%) samples, while the highest rate was found in the Netherlands (12/26, 46.2%).

A total of 132 open-text answers were provided about the use of CDSS, and 49 answers to the question on predictive CDSS, of which 105 (79.5%) and 38 (77.6%) were analyzed, respectively. For the general CDSS question, all the exclusions stemmed from the answer being too unclear and therefore not being classifiable. Such answers, for instance, were not specific enough (eg, “I use multiple algorithms”) or did not clearly mention a CDSS (eg, “thromboembolism”, “standard procedures”). For the predictive CDSS question, answers that concerned nonpredictive CDSS were also excluded (eg, “CORNELL”, a diagnostic instrument).

As can be seen in Figure 1, multiple dominant categories of CDSS were mentioned, such as diagnostic instruments (24/105, 23%), prediction tools (20/105, 19%), and tools considering intervention advice (17/105, 16%). Examples of these categories include the mini mental state examination or WELLS criteria (diagnostic instruments), CDSS on care pathways or future bleeding risks (prediction tools), and interaction checkers and antibiotic prescription systems (intervention advice). Another common category mentioned by professionals was geriatric assessment tools (16/105, 15%), most of which were considered interRAI instruments. Finally, some professionals mentioned commercial software solutions (N=2) providing one or more CDSS categories, but these responses lacked sufficient detail to identify specific categories or methods and were therefore grouped into a separate cluster.

When looking at the types of CDSS, most tools were knowledge-based (71/105, 68%) and aimed for diagnostic (69/105, 65%) purposes, thereby making predictive or prognostic tools much less common in current practice, only mentioned in about 19% (20/105) of the answers to the general CDSS question.

While analyzing the predictive CDSS mentioned, 24 (49%) answers had to be excluded, as they did not concern any predictive tool. Out of the remaining 26 valid predictive CDSS answers, most concerned either intervention advice (eg, CHA2DS2-VASc or CVRM treatment advice) or predictive tools such as (cardiovascular) mortality risk calculators. Some of the overall CDSS categories were excluded, as they are not predictive by definition. Note that in Figure 1, “prediction models” includes all CDSS providing direct outcome predictions based on responses to the general CDSS question.

Tables 3 and 4 show the results of the regression models built to explain being comfortable with using technology, and for the use of CDSS. In the first model, age and openness to new technologies were significant determinants of feeling comfortable with technology. Age showed an inverse association (β=–0.022; P<.001), while people who are more open to technology showed a significant positive association (0.622; P<.001). In the second model, no determinants were significantly associated with the actual use of CDSS.

A total of 136 (90.1%) participants answered the question on what they perceive as future beneficial CDSS s in their clinical work (“What kind of clinical decision support tool would be beneficial for you in patient care?”), out of which we retrieved 179 valid answers, as some answers contained multiple technologies. These answers were qualitatively clustered in seven categories in total with individual subcodes as subcategories, as can be found in Table 5.

Frequently mentioned were systems that would either estimate risks and outcomes for patients, such as providing information on risk of mortality, hospitalization, or falls. Next, many professionals also mentioned that support systems giving advice on certain interventions (either pharmacological or nonpharmacological) could also be of help in their practice. Often, professionals would describe the need for personalized information on, for instance, risks for side effects or the effectiveness of the chosen intervention. These 2 categories, although depending on the exact use case, concern tools that would provide clinicians with prognostic information on treatment advice and important health outcomes. Especially, doctors seemed to be interested in such CDSS. Nurses were interested in these too, but more often reported future beneficial technology from the “other” category. We saw only a tendency toward more reporting of recommendation systems in the home care sample.

An important part of the answers also included the need for either electronic record adoption or improvements within an electronic record already in use. New functionalities such as automated note writing based on the recorded consultation or an integrated sharing between professionals were often mentioned. Finally, although generative AI was only mentioned once in the current CDSS use, multiple respondents answered that they would benefit from a clinical reasoning tool that would come up with differential diagnosis or with which they can discuss complex patient cases.

This study provided an overview of the current use of CDSS by health professionals working with older persons with CCC from six European countries and the United States. The study highlighted notable variations in use rates and professionals’ needs for future developments in CDSS. The results also revealed differences in CDSS adoption. For example, adoption rates were as low as 15% (3/20) in the Italian sample, contrasted by remarkably high use at 90% (18/20) in the Finnish sample. These disparities are possibly influenced by varying degrees of investment in digital infrastructure, policy support mechanisms, and training, along with cultural attitudes to adopting technology in clinical practice [21,22]. In our study, most participants reported being open to new technology and comfortable using new technology, although the overall uptake of CDSS was 51.7% (78/151), and of predictive CDSS only 22.5% (34/151).

These results are similar to previous research indicating that professionals will adopt CDSS if they perceive benefits and are supported to implement the tools [23,24]. As a general principle, an effective CDSS must minimize the effort required by clinicians to receive and act on system recommendations [1]. However, as both a limitation of this study and a topic for reflection, we found a considerable proportion of unsatisfactory responses to questions regarding CDSS, particularly those addressing their predictive functionalities. This finding may indicate a lack of understanding among participants concerning the nature and capabilities of predictive CDSS. Prior studies have similarly identified a gap in healthcare professionals’ literacy on such topics, emphasizing the importance of addressing this issue to facilitate the effective implementation of CDSS in clinical practice [25-27].

Moreover, this study did not evaluate the performance of individual CDSS, which can be a topic of further research. Notably, certain reported CDSS (eg, generative AI) are not (yet) recommended for clinical practice. When combined with limited literacy among clinicians, the adoption of such tools could introduce risks to patient safety and care quality.

The findings from the study showed cross-national variations for comfort and openness to technology, and variation in the actual use of CDSS and predictions. Our findings showed that, although clinicians were comfortable or open to technology, this seldom correlated with the reported use, suggesting that although clinicians are keen to adopt such tools, there may remain infrastructural or systemic obstacles to real adoption. The results from the regression model for CDSS use are congruent with this. Previous research has also shown that differences in micro and macro-level factors across countries, for example, individual professionals’ views, costs, or national legal issues, can influence the adoption [28].

In addition, the results indicated that there is an important need among health care providers for improvements in CDSS. As most of our responses concerned either personalized risk assessments or general diagnostic or intervention recommendations, there is a practice-driven demand for further research improving healthcare prognostications on a personalized level [28]. This could, for instance, include personalized treatment advice [29]. This pilot was part of a project aiming at advancing the field on this subject, but as our data shows, there are many more outcomes or interventions to be studied.

Interestingly, consistent with previous research, we found a significant number of existing CDSS tools, even some of which could fill the needs mentioned as future beneficial CDSS, it seems CDSS are only used to a limited extent [30]. It has indeed been previously described that CDSS are often developed, but they often encounter challenges in achieving sustained implementation and long-term integration into practice [31-33]. This gap between development and real-world adoption highlights critical challenges in translating innovation into practice, such as integration with existing health IT systems, clinician training, workflow adaptation, and organizational readiness [34].

We aimed to include participants with diverse paramedical backgrounds and from different countries, providing an overview across multiple countries and career stages. As only about 20 participants were included per country, the data cannot be regarded as representing the countries as a whole. Furthermore, the participant recruitment was, as described in the pilot protocol, based on targeted purposive samples and might therefore be biased toward research and innovation-oriented professionals and professionals with interRAI experience. However, it can be argued that for parts of the results, such as the future beneficial CDSS, we still reached a representative overview of what professionals could benefit from. Finally, as we worked with open-ended questions, the response categorization could still be influenced by subjective interpretations despite working with coding procedures. While we tried to retrieve additional information on the answered CDSS where possible, asking participants additional questions on the mentioned systems could have further refined the analysis. The responses to the questions also depend on the respondents’ understanding of the question. Although a trained facilitator was always available for any questions or clarifications, we found that not all answers fitted the original questions, mostly due to lack of knowledge or misinterpretation.

This study collected the experiences with CDSS of healthcare professionals in a purposive sample across Europe and the United States. We found that CDSS use varied across countries samples, with a limited uptake in general, and particularly, for predictive CDSS. Participants mentioned multiple examples of CDSS they used, mostly for diagnostic rather than predictive purposes. Nevertheless, many professionals reported that predictive tools to inform them about patients’ health and intervention outcomes would be most beneficial. This highlights the need for further implementation of current systems and the development of more predictive CDSS.