Clinical decision support systems (CDSS) can improve guideline adherence and reduce variability. HTE-DLP 3.0 (Virtual Lipidologist) provides recommendations based on scientific evidence, safety, and cost-effectiveness criteria.
MethodsA proof-of-concept study was conducted using 10 standardised clinical cases evaluated by 9 clinicians before and after the use of the HTE-DLP 3.0. Therapeutic variability was analysed using the Simpson Diversity Index and the kappa coefficient. Usability and user experience were assessed using the CSUQ and QoE questionnaires, together with a qualitative survey on strengths and areas for improvement.
ResultsBaseline assessment showed very significant interprofessional variability. Uniform adherence to the shared algorithmic recommendation is observed. The CSUQ showed good efficiency scores (6.40) and high satisfaction (5.42/7), with error messages and the availability of clinical support tools identified as areas for improvement. The QoE showed high scores for data security (4.71), clinical usefulness (4.57), and social impact (4.57), with learning time being the lowest-rated aspect (2.85/5).
ConclusionsHTE-DLP 3.0 demonstrated the ability to act as a decision standardization tool based on clinical practice guidelines. Further studies in real-world clinical practice will be necessary to evaluate its clinical and cost-effectiveness impact.
Los sistemas de ayuda a la decisión clínica (CDSS) pueden mejorar la adherencia a las guías y reducir la variabilidad. HTE-DLP 3.0, Lipidólogo Virtual, proporciona recomendaciones basadas en evidencia científica, seguridad y criterios de coste-efectividad.
MétodosPrueba de concepto utilizando 10 casos clínicos estandarizados evaluados por 9 clínicos antes y después del uso de HTE-DLP 3.0. La variabilidad terapéutica se analizó mediante el índice de diversidad de Simpson y el coeficiente kappa. La usabilidad y experiencia de usuario se evaluaron con los cuestionarios CSUQ y QoE, junto con una encuesta cualitativa sobre fortalezas y áreas de mejora.
ResultadosEn la evaluación basal se observó una variabilidad interprofesional muy significativa. Se objetiva adhesión uniforme a la recomendación algorítmica común. El CSUQ mostró buenas puntuaciones de eficiencia (6,40) y alta satisfacción (5,42/7), identificándose como áreas de mejora los mensajes de error y la disponibilidad de herramientas de apoyo en el manejo clínco. El QoE mostró altas puntuaciones en seguridad de datos (4,71), utilidad clínica (4,57) e impacto social (4,57), siendo el tiempo de aprendizaje el aspecto peor valorado (2,85/5).
ConclusionesHTE-DLP 3.0 mostró capacidad para actuar como una herramienta de estandarización de la toma de decisiones en base a las guías de práctica clínica. Serán necesarios estudios en práctica clínica real para evaluar su impacto clínico y coste-efectivo.
Hypercholesterolaemia is a causal factor in atherosclerotic disease, responsible for one-third of global morbidity and mortality.1 The Santorini-Spain study showed that, after one year of follow-up, only 34% of patients at high or very high vascular risk achieved the lipid targets recommended in the 2019 European guidelines, further demonstrating an underutilisation of the available therapeutic options.2 The 2025 update of the European Society of Cardiology (ESC) and European Atherosclerosis Society (EAS) clinical guidelines establishes a treat-to-target approach, prioritising both percentage reductions and the achievement of absolute LDL cholesterol levels according to the patient's cardiovascular risk.3
However, numerous studies have documented a significant gap between guideline recommendations and actual clinical practice, particularly regarding the appropriateness of the prescribed treatment potency. In an analysis conducted in primary care in Spain, only 13.2% of patients with an indication for lipid-lowering therapy received an intensity appropriate to their cardiovascular risk as determined by the 2019 ESC/EAS guidelines, and 74% did not reach their targets after 3 years of follow-up.4 A study of more than 900,000 patients showed that up to 30% of cases involve the use of statins metabolised by the CYP3A4 pathway concomitantly with potent inhibitors of this enzyme, including approximately 9% of patients exposed to an inhibitor with an explicit contraindication in the product information, without changing their treatment to statins not metabolised by this pathway (such as rosuvastatin, pravastatin, or fluvastatin). This practice could increase the prevalence of adverse effects, reduce treatment tolerance, and highlights the need for personalised therapy.5 Several studies have shown the difficulty in clinical practice of accurately identifying vascular risk, with accuracy rates around 60%.6
The Cardio Right Care Cardiovascular Risk Control Project, conducted with the participation of more than 900 Spanish professionals from both primary care and cardiology, revealed wide variability in adherence to different clinical practice guidelines and vascular risk scales, showing that 12% of participants did not follow any guidelines.7 A recent study showed that it takes approximately 17 years for the scientific evidence compiled in guidelines to be widely incorporated into clinical practice.8 Taken together, these data reflect a clear and urgent need for tools to support the clinical management of cardiovascular risk factors, especially in the field of dyslipidaemia.
Clinical decision support systems (CDSS) have been proposed as effective tools to improve adherence to guidelines, facilitate risk stratification, and recommend personalised treatments in real time.9 To maximise their impact, it is essential that they be integrated into the electronic health record, activated automatically, and generate actionable, specific, and relevant recommendations for the patient.10 Despite their potential, the implementation of CDSS remains limited. Recent studies demonstrate that their effectiveness depends critically on design, usability, and acceptance by healthcare professionals.11
A systematic review of CDSS focused on lipid-lowering treatment, which included 34 studies and 87,874 patients, showed an improvement in lipid levels and cardiovascular outcomes.12
In this context, our group, in collaboration with the Catalan Lipid Unit Network (XULA), developed CDSS HTE-DLP 2.0, the first specific system for lipid-lowering treatment and the management of familial hypercholesterolaemia (FH) validated at the national and European levels.13 It demonstrated a fourfold increase in the number of patients achieving lipid targets, an estimated 6% decrease in the incidence of coronary heart disease, greater confidence in treatment selection,13 estimated healthcare savings of €33 million at the national level,14 and good acceptance by clinicians.15 HTE-DLP 2.0 received the certificate of educational interest from the Spanish Society for Medical Education.16 Subsequently, HTE-DLP 3.0 was developed, which included various specific tools and scales for managing patients with FH. Recently, the ITEMAS platform of the Carlos III Health Institute has included HTE-DLP 3.0 in its service portfolio due to its high potential for transfer to the National Health System.17
This study aims to evaluate the capacity of an algorithmic recommendation to reduce prescribing variability among expert clinicians, assess the user experience, and lay the groundwork for the co-creation of a future version.
MethodologyHTE-DLP 3.0 is a CDSS that replicates, using closed Boolean algorithms, the decision-making process a lipid specialist would undertake to personalise lipid-lowering treatment based on effectiveness, safety, and cost-effectiveness criteria. It incorporates the clinical practice guidelines that should be considered when choosing the optimal treatment.18–27 (Table 1). The system has the capacity for automated detection of patients with a lipid profile indicative of genetic dyslipidaemia and to alert healthcare professionals, offering support for management and treatment planning. The healthcare professional must manually enter the lipid values and select from the different options offered by the system semi-automatically.
Clinical decision structure of the HTE-DLP 3.0 Virtual Lipidologist.
| Clinical decision block | Key question | Incorporated guidelines and documents |
|---|---|---|
| 1. Risk stratification | What is this patient's cardiovascular risk? | ESC/EAS Guidelines 201918 SEA Standards 202221 |
| 2. Diagnosis of FH | Does the patient have a lipid profile indicative of familial hypercholesterolaemia (FH)? What is their DLCN score? Are genetic testing and cascade screening indicated? | Dutch Lipid Clinics Network Criteria; ESC/EAS 201918 |
| 3. Therapeutic goals | What LDL-C targets are appropriate based on their cardiovascular risk? | ESC/EAS 201918; Masana and Plana Therapeutic Planning Tables19 |
| 4. Lifestyle | What lifestyle recommendations are most suitable for this patient? | SEA Document 201820 |
| 5. Treatment intensity | What percentage reduction in LDL-C is necessary? | Masana and Plana Therapeutic Planning Tables19 |
| 6. Drug selection | What lipid-lowering treatment (monotherapy or combination therapy) will allow them to reach the target? | ESC/EAS Guidelines 201918 |
| 7. Safety and interactions | Are there any relevant drug interactions or contraindications? | Structured Drug Interaction Review22 |
| 8. Comorbidities | Are there any comorbidities that limit treatment options or dosages (e.g., chronic kidney disease)? | Drug Adjustment in CKD23 |
| 9. Cost-effectiveness | If efficacy and safety are equal, which option is more cost-effective? | National Health System Reference Pricing System24 |
| 10. Funding | Are there any funding criteria or therapeutic positioning guidelines? | Alirocumab IPT25 |
| 11. Follow-up | How often should clinical and laboratory follow-up be performed? | Evolocumab IPT26 |
| 12. Long-term safety | How can adverse effects of the treatment be detected and managed? | Review of statin adverse effects27 |
List of sequential questions and documents incorporated in HTE-DLP 3.0. This table reflects the sequential decision-making process that a clinical lipidologist expert theoretically follows in routine practice to personalise lipid-lowering treatment based on clinical effectiveness and safety criteria, and, all other things being equal, according to criteria for minimising direct costs. It also includes the scales and tools that should be part of the "expert lipidologist's toolkit.".
Version HTE-DLP 3.0 includes statins, ezetimibe, and proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitors as therapeutic options. Access to the multiplatform website is online, after user authentication. The server is hosted within the hospital information system and protected under the standard cybersecurity regulations for healthcare information established by the Department of Health (Fig. 1). Link to the application: https://hte.salutms.cat:3000/#/patient/fill_patient_data.
The HTE-DLP 3.0 version presents the following differences compared to the previous version, HTE-DLP 2.0: it is based on the 2019 ESC/EAS guidelines for the treatment of dyslipidaemias, whereas the previous version was based on the 2011 EAS/ESC guidelines; it has the capacity to be integrated into the hospital information system; it includes a specific module for managing patients with familial hypercholesterolaemia (FH) with automated detection of patients with a phenotype indicative of FH; it provides support materials, references to websites, and validated information for professionals and patients; and it features a communication system with reference lipid units.
The study design was a pre-post intervention proof-of-concept study, using the same simulated clinical cases in both evaluation rounds, with the participation of 9 lipid specialists from 6 lipid units accredited by the Spanish Society of Arteriosclerosis. This approach, common in pilot and proof-of-concept studies, allows for the exploration of variability and convergence patterns in clinical decision-making, as well as the system's usability, in a controlled environment that limits its external validation. However, it can introduce learning or recall bias, which must be considered when interpreting the results.
Experts anonymously resolved 10 simulated clinical cases (primary prevention in patients with familial hypercholesterolemia in children and adults, secondary prevention, patients with comorbidities such as chronic kidney disease, and polymedicated patients with a high risk of drug interactions). These cases were validated by two vascular risk experts (the cases are listed in Appendix 1). Two consecutive anonymous rounds were conducted one week apart, with three possible responses per clinical case: R1 without system support and R2 using HTE-DLP 3.0.
Inter-prescriber variability and agreement with the system's recommendations were analysed using Cohen's kappa coefficient.28 A total of 270 responses were analysed per round. Quantitative data were analysed using R v4.3.2.
Participants completed standardised questionnaires: the Computer Systems Usability Questionnaire (CSUQ)29 and the Quality of Experience (QoE) Questionnaire for mHealth applications.30 Qualitative data were obtained through an open-ended questionnaire on strengths, limitations, and recommendations for improvement.
The study was conducted in accordance with the WHO ethical code (Declaration of Helsinki) for experiments and was based exclusively on simulated cases, without the participation of real patients.
ResultsThe participants were 4 women, with a mean age of 40.28 years (SD 23.78), and 5 men, with a mean age of 54.20 years (SD 14.20). The mean professional experience was 24 years (SD 4.69) for women and 25.4 years (SD 14.4) for men.
The interprofessional agreement coefficient in the selection of lipid-lowering treatment was κ = .25. The range of κ varied between .00 and .33, with no pair reaching the threshold of moderate agreement. These values are considered poor or low (Fig. 2). Cases 3, 4, 7, and 9, with κ = .0556, showed the greatest dispersion in drug selection. In all of these cases, it was necessary to choose the lipid-lowering treatment with the lowest risk of drug interactions, given that these were patients on multiple medications or at high risk due to renal insufficiency. The highest agreement was observed in clinical case 2, corresponding to a 35-year-old woman with heterozygous FH and baseline LDL cholesterol levels of 120 mg/dL, treated with atorvastatin 20 mg and ezetimibe 10 mg (Table 2). The use of HTE-DLP 3.0 reduced variability by 53%.
Detailed Results of the Computer Systems Usability Questionnaire.
| Item CSUQ | Domain | Mean (SD) | Interpretation |
|---|---|---|---|
| I am satisfied with the ease of use of the platform. | Usability | 5.28 (1.11) | Good overall satisfaction with ease of use |
| It has been simple to use the platform. | Usability | 5.00 (1.41) | Perceived as easy to use by most |
| I can complete my work quickly. | Efficiency | 6.40 (1.14) | Excellent efficiency and speed |
| I feel comfortable using the platform. | User Experience | 5.00 (1.41) | Comfortable use and positive experience |
| It has been easy to learn how to use the platform. | Learning | 5.00 (1.00) | Good accessibility and appropriate learning curve |
| I have improved my clinical skills. | Clinical Utility | 4.71 (1.38) | Contributes to improving competencies in dyslipidaemia |
| The error messages clearly indicate how to resolve problems. | Feedback | 4.14 (1.46) | Needs improvement in the clarity of error messages |
| When I have made a mistake, I have been able to fix it quickly. | Feedback | 4.57 (1.60) | Adequate ability to resolve errors |
| The information provided is clear. It is easy to find the information I need. | Information | 5.00 (1.41) | Information perceived as clear and useful |
| The information has been effective in completing tasks. | Information | 4.85 (1.34) | Good accessibility to information |
| The organization of the information is clear. | Information | 4.85 (1.21) | Effective information for practice |
| The interface is user-friendly. | Design | 4.85 (1.21) | Appropriate visual organisation |
| I have enjoyed using the platform. | Design | 4.71 (1.70) | Well-rated interface |
| The platform has all the tools I expected. | Overall Satisfaction | 5.42 (1.39) | High overall satisfaction |
| I am satisfied with the platform. | Functionality | 4.00 (1.53) | Some expected tools are missing |
| I would recommend the platform to other professionals. | Overall Satisfaction | 4.85 (1.34) | Good overall satisfaction |
| I am satisfied with the ease of use of the platform. | Overall Satisfaction | 5.00 (1.26) | High likelihood of recommendation |
The results obtained (mean ± SD) from the Computer Systems Usability Questionnaire (CSUQ) and their interpretation are shown. This questionnaire evaluates the user's subjective satisfaction with a computer system, analysing aspects such as ease of use, efficiency, and effectiveness in performing tasks. Its purpose is to obtain direct information from users about their impressions and experience with the system. The maximum score for each item is 7.
In the analysis of the 10 simulated clinical cases, the degree of agreement with the system's recommendations reached 68%. The use of combinations with ezetimibe and/or PCSK9 inhibitors increased from 42% to 71% after using HTE-DLP 3.0. After using HTE-DLP 3.0 (R2), all experts adopted the therapeutic recommendation generated by the system for each clinical case, resulting in complete convergence toward a single therapeutic option per scenario. This agreement reflects uniform adherence to a common algorithmic recommendation and should not be interpreted as an independent clinical consensus among experts. In this context, the kappa coefficient loses interpretive power, since the absence of interobserver variability prevents estimating agreement beyond chance.
The mean overall CSUQ score was 4.91/7 (SD 0.5) and the mean QoE score was 4.13/5 (SD 0.4). Detailed results are shown in Tables 2 and 3.
Results of the Quality of Experience Questionnaire for mHealth Applications.
| Item QoE (translation) | Domain | Mean (SD) | Interpretation |
|---|---|---|---|
| The app's calculations are correct. | Technical reliability | 4.00 (0.58) | Good confidence in the accuracy of the calculations |
| The traditional method is more difficult. | Comparative usability | 3.28 (0.95) | Slight preference for the app over the traditional method |
| I found what I needed. | Access to information | 3.14 (0.90) | Moderate accessibility; room for improvement |
| The app is useful for monitoring the disease. | Clinical utility | 3.85 (0.69) | Provides real support for clinical monitoring |
| It provides the features I expected. | Functionality | 3.28 (0.95) | Some expected features are missing |
| The data provided is reliable. | Reliability | 4.57 (0.49) | High perception of reliability |
| It receives regular updates. | Maintenance | 3.28 (0.90) | Low visible maintenance is perceived |
| It allows me to send information to my doctor. | Connectivity | 3.00 (1.00) | Feature is underutilised or difficult to find |
| My work quality improves with the app. | Workplace impact | 3.42 (0.79) | Moderate improvement in professional performance |
| I identify with the health issues addressed. | Clinical relevance | 3.85 (0.69) | Adequate clinical relevance of the content |
| The learning curve is adequate. | Usability/learning | 3.42 (0.79) | Learning curve could be improved |
| I have guaranteed access to my data at all times. | Security/accessibility | 3.57 (0.79) | Acceptable, but not optimal, level |
| It has adequate security methods. | Security | 4.00 (0.58) | Good perception of technical security |
| The data is sufficiently protected. | Safety | 4.00 (0.58) | Protection is perceived as adequate |
| The design is adequate. | Design | 3.85 (0.35) | Design is well-received |
| I would change or add something. | Functionality | 3.71 (0.49) | Users request functional improvements |
| Performance could be further optimised. | Technical performance | 2.85 (1.07) | Clear room for optimisation is perceived |
| I encountered errors while using the app. | Stability | 3.00 (0.58) | Errors are occasionally detected |
| I would use the app if it were developed. | Acceptability | 4.42 (0.53) | High intention to use it in the future |
| It would aid in the treatment of diseases. | Clinical value | 4.57 (0.49) | High perceived clinical value |
| It would be useful for society. | Social impact | 4.57 (0.49) | High potential for social impact |
| It would improve the user's quality of life. | Health impact | 4.71 (0.49) | Very high potential to improve patients' lives |
The results obtained (mean ± SD) on the Quality of Experience Questionnaire for mHealth Applications (QoE) are shown. This scale measures the user's subjective perception of the quality of a mobile health application, considering overall satisfaction, usability, and other factors that influence the overall experience. The maximum score for each item is 5.
The qualitative evaluation of the tool revealed strengths, highlighting adherence to clinical guidelines, cost-effective prioritisation, automated risk calculation, and the perceived safety and clinical utility. However, clear areas for improvement were identified, particularly in the automated entry of relevant clinical data, the detection of drug interactions, the full integration of diagnostic scales, the optimisation of the mobile device experience, and the available toolset. Professionals emphasised the need to adapt the platform to regional administrative guidelines, improve error messages, and enhance learning and navigation features. Furthermore, the incorporation of artificial intelligence systems to streamline and automate various functions is recommended (Table 4).
Qualitative comments on the evaluation of HTE-DLP 3.0. Strengths, weaknesses and suggestions for improvement.
| Strengths | Weaknesses | Suggestions for improvement |
|---|---|---|
| Methodological rigor | Lack of automated drug interaction detection | Add new features requested by the clinical team (e.g., automated capture of drug interactions) |
| Cost-effective prioritszation | Incomplete integration of diagnostic scales | Optimise the alert and error message system |
| Automation of risk calculation | Limitations in mobile device optimisation | Improve tutorials, onboarding, and guides within the platform |
| Perception of safety and clinical utility | Lack of fatigue assessment via alerts | Expand input fields (triglycerides, total cholesterol, HDL-C, Lp(a), contraindications, glomerular filtration rate, etc.) |
| Transparency in findings | Lack of fully automated parameter capture | Update the system with the full spectrum of currently marketed lipid-lowering drugs |
| Adherence to clinical guidelines | Missing tools expected by healthcare professionals (e.g., local administrative guidelines) | Rank therapeutic options by LDC-C interval rather than absolute value |
| High efficiency and speed in the care process | Insufficient error messages and feedback | Prioritise options based on cost criteria |
| Good perceived clinical utility | Considered steep learning curve | Include specific state and regional administrative recommendations |
| Perception of safety and reliability of the information. | Occasional difficulty entering patient-specific data | Implement AI systems |
Subjective comments on HTE-DLP 3.0 are shown, evaluating its strengths (internal positive attributes), weaknesses (internal limitations) and suggestions for improvement expressed by the participants to guide the design and development of future versions of HTE-DLP Virtual Lipidologist.
This evaluation of the HTE-DLP 3.0 system demonstrates that a specialised dyslipidaemia CDSS, based on deterministic algorithms and explicit rules, can significantly improve adherence to clinical practice guidelines, standardise clinical decisions, and facilitate treatment intensification in accordance with current recommendations.
The baseline variability observed among experts, even in standardised cases, is consistent with previous studies describing significant inconsistencies in the selection of lipid-lowering treatments, both in primary care and in specialised settings.31,32 These discrepancies are related to multiple factors, such as the diversity of available guidelines, differences in individual risk perception, and healthcare pressures, which hinder a comprehensive assessment. HTE-DLP 3.0 demonstrated the ability to reduce this variability by more than 50%, providing consistent recommendations based on explicit criteria of efficacy, safety, and cost-effectiveness. This reduction is consistent with the results of other cardiovascular CDSSs, which have been shown to improve guideline adherence and promote appropriate intensification strategies.33 The apparent discrepancy between the complete convergence observed after the intervention and the low kappa coefficient values highlights an inherent limitation of agreement metrics when applied to forced-decision scenarios. The use of a CDSS that provides a single recommendation per case eliminates interobserver variability, thus limiting the applicability of kappa as a measure of agreement. For this reason, the results should be interpreted as a marked reduction in interprofessional variability through algorithmic homogenisation, and not as a demonstration of spontaneous clinical agreement.
The use of HTE-DLP 3.0 was associated with a higher prescription rate for combination therapy with ezetimibe and PCSK9 inhibitors, increasing from 42% to 71%, and therefore with a higher intensity of lipid-lowering treatment. Intensive LDL cholesterol reduction using a combination of statins and non-statin therapies is a cornerstone of treatment for patients at high or very high cardiovascular risk, as it halts the progression and promotes the regression of atherosclerotic plaques; therefore, it should be considered a standard therapeutic approach from the start of treatment.34 The increased use of combination therapies with the HTE-DLP 3.0 system reflects the strict application of clinical guidelines integrated into the system's algorithm, especially in very high cardiovascular risk or heart failure scenarios. However, this finding should be interpreted with caution, as a uniform intensification of therapy may carry a potential risk of overtreatment in certain clinical profiles, particularly in primary prevention. These results underscore the need for clinical decision support systems (CDSS) to act as supportive tools and not as substitutes for individual clinical judgment, as mandated by law.
Several studies have demonstrated the usefulness of specific lipid management systems (CDSS), both in dyslipidaemic patients in general35 and in patients with familial hypercholesterolaemia (FH),36 among which HTE-DLP 2.0 stands out, quadrupling the number of patients achieving lipid targets.13 Another key finding is the improvement in the perception of clinical utility, efficiency, and safety, reflected in the overall scores of the CSUQ and QoE questionnaires. These findings are consistent with previous studies that emphasize that the adoption of CDSS depends critically on usability, interface clarity, cognitive load, and the availability of relevant information in real time.37 Unlike other existing CDSS, which are often limited to single-focus recommendations or based solely on risk, HTE-DLP 3.0 adopts a holistic approach by integrating guidelines, drug safety, territorial funding criteria, and cost-effective prioritisation, aligning with current trends in intelligent systems focused on value and equity.
However, several areas for improvement were identified. Professionals noted limitations primarily related to user learning, system feedback, and data integration. The lowest-rated aspect of Quality of Experience (QoE) was the time required to learn how to use the application, highlighting the need to improve support guides and visual aids. Similarly, the insufficient clarity of error messages underscored the need to optimise alerts to facilitate incident resolution. Regarding functionality, clinicians highlighted shortcomings such as the lack of expected tools for automating the entry of relevant data (complete lipid profile, concomitant treatments, renal function), as well as the need to fully integrate automated diagnostic scales and update therapeutic options to include bempedoic acid and inclisiran. Although the design was well-received, improvements to technical performance and the correction of specific errors were recommended. Finally, the most critical aspect identified was the full integration and automation of data capture, repeatedly cited as the main challenge for optimising the usability and clinical impact of CDSS, a limitation also described in recent evaluations of digital tools in cardiometabolism.38
Despite the potential of CDSS to improve patient care and cardiovascular health outcomes, several barriers persist that hinder their effective implementation, limiting their ability to achieve health goals. Among the main barriers are time and resource constraints, alert fatigue, technical difficulties with the system, lack of trust in the recommendations, and the inherent complexity of managing cardiovascular disease in real-world practice.39 Co-creation of CDSS with the participation of clinicians and patients from the design phase and throughout the continuous improvement cycle is key to achieving successful implementation in clinical practice. Scientific societies can and should play a fundamental role in leading the development, validation, implementation, and monitoring of these tools in healthcare settings.
This study has several limitations. First, the consecutive pre-post design with the same clinical cases may introduce a learning bias, although limited given the small number of cases. Second, the small number of participants and the exclusive inclusion of lipid experts limit the validity of the results in routine clinical practice, although it is expected to be useful if used by non-expert professionals. Third, the use of simulated cases prevents the evaluation of real clinical outcomes, although these were analysed in the previous version (HTE-DLP 2.0). The results are useful for decision patterns in controlled settings, which limits their validity in external settings. Finally, the observed convergence reflects algorithmic homogenisation and does not allow for inferring a direct improvement in individual clinical appropriateness.
Future versions should evolve into a CDSS with automatic data capture, based on artificial intelligence, and with the capacity for periodic and autonomous updates.40 Incorporating this architecture could allow the integration of the lipid axis with other vascular risk domains, fostering a more comprehensive, proactive, and personalised approach to cardiovascular risk.
ConclusionsHTE-DLP 3.0 demonstrated its ability to standardise decision-making and improve adherence to clinical practice guidelines in controlled environments and under clinical supervision. The observed therapeutic intensification reflects the strict application of guideline-based recommendations but underscores the need to maintain individual clinical judgment. Future studies with new versions incorporating the identified improvements, with larger sample sizes, participation from diverse clinical profiles, and evaluation in real-world clinical practice will be necessary to determine the clinical and cost-effective impact of these tools.
CRediT authorship contribution statementAZ is the principal study researcher. They conceived the original idea, led the conceptual design of the project, and co-developed the HTE-DLP 3.0 clinical decision support tool in close collaboration with the technical team. They also actively participated in defining the study protocol, overseeing data collection, analysing and interpreting the results, and drafting the manuscript.
LM, along with AZ, coordinated the study's methodological design, contributed to the data analysis, and played a significant role in drafting and critically reviewing the manuscript.
The remaining authors contributed to the study's implementation and the clinical validation of the tool at their respective centers, participating in the completion of surveys, the clinical evaluation of results, and the interpretation of findings. They also critically reviewed the manuscript, providing substantial clinical and scientific feedback.
All authors approved the final version of the manuscript and accept responsibility for the integrity and accuracy of the work presented.
Declaration of Generative AI and AI-assisted technologies in the writing processDuring the preparation of this study, the authors used ChatGPT 5.0 in the writing process to improve readability and written language. After using this tool, the authors reviewed and edited the content as needed, assuming full responsibility for the content of the publication.
FundingThis study was funded by the Clinical-Epidemiological Grant in Arteriosclerosis from the Spanish Society of Arteriosclerosis, 2022; the Research Grant from the Spanish Foundation of Internal Medicine, Spanish Society of Internal Medicine, 2024; and the Rubiés Prat Grant 2024, from the Catalan Network of Lipid and Arteriosclerosis Units. Spanish Society of Healthcare Managers Foundation (SEDISA) Research Grant, 2024.
The authors declare no financial or personal conflicts of interest with other individuals or organisations related to the article submitted for publication.
To the ICT Department of the Maresme i la Selva Health Corporation, especially Óscar Rivas, for their involvement in the technological development of the project.
To Laura Muñoz, from Datexbio, for her support in the statistical analysis.
