We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
The Aleph: A Multi-Purpose Clinical Decision Support Platform for Palliative Care Screening.
- Authors
Doñate-Martínez, Ascensión; Blanes-Selva, Vicent; Linklater, Gordon; Asensio-Cuesta, Sabina; Pereira Mesquita, Felipe; Garcés, Jorge; Sánchez-García, Ángel; García-Gómez, Juan M.
- Abstract
Introduction: A significant amount of research on palliative care (PC) has demonstrated the benefits of identification for a timely referral as well as enabling an integrated care approach. However, the prompt identification of PC in clinical practice has been a challenge. Despite PC needs may differ widely between patients, a short survival prediction is one of the most common proxies to start PC conversations, therefore predicting if the death will occur within the next year has a real value when screening for patients who may benefit from PC. Our objective was to create a clinical decision support system called Aleph to support healthcare providers in their screening for patients with PC needs. Aleph incorporates different quantitative and data-driven predictive services based on machine learning technology and electronic health records (EHR) data. Methods: Two machine learning services have been developed using retrospective hospital admission data from the EHR (39,310 episodes from 19,753 different patients; age >= 65). The first service incorporated three models: t models: a one-year mortality classifier, a mortality regressor, and a one-year frailty status classifier. The second service incorporated a compact and minimalist version of the one-year mortality models using only common variables. We validated Aleph's usability and user experience using a user-centred methodology involving 21 healthcare providers from six countries. Results: The models on the first service reported an area under the ROC curve (AUC-ROC) of 0.87 for the one-year mortality classifier, 94.67 days of mean absolute error for the regression model, and 0.89 of AUC-ROC for the one-year frailty classifier. The compact model on the second service reported an AUC-ROC of 0.83. Aleph made a great impression among the validation participants and scored 65 over 100 points on System Usability Scale and 1.5 over 3 in the User Experience Questionnaire-Short. Discussion: Aleph was developed as a multi-purpose clinical decision support system web application. Its design allows the incorporation of further integrated care services, even out of the realm of the PC. We have incorporated two PC services based on powerful predictive models. Documentation to create a new predicted service will be openly distributed to encourage researchers to contribute to the Aleph so their predictive models could be incorporated immediately into a usability-validated platform. Conclusion: Our contribution to the identification of patients who may benefit from PC are two predictive services with high predictive power models and a user-centred validated platform. These allow the integration of this outcome with other services available to address patients' needs as well as to 'plug and play' other predictive services. Implications for applicability and limitations: The use of Aleph can improve the detection of patients in need of PC, helping physicians and hospital administrators to address the growing demand for PC and improving the patient's care. The main limitation of the study is the lack of validation of the predictive model with other data sources.
- Subjects
DENMARK; MEDICAL screening; MACHINE learning; CONFERENCES &; conventions; NEEDS assessment; DECISION making in clinical medicine; SYSTEMS development; PALLIATIVE treatment
- Publication
International Journal of Integrated Care (IJIC), 2022, Vol 22, p1
- ISSN
1568-4156
- Publication type
Article
- DOI
10.5334/ijic.ICIC22372