Introduction: We developed a machine learning algorithm to predict the risk of emergency hospitalization within the new 7 to 14 days with a good predictive performance (AUC=0.85). Data recorded by home aides were send in real time to a secure server to be analyzed by our machine learning algorithm, which predicted risk level and displayed it on a secure web-based medical device. This study aims to implement and to evaluate the sensitivity and specificity's predictions of Presage system for four clinical situations with a high impact on unscheduled hospitalization of older adults living at home: falls, risk of depression (is sadder), risk of undernutrition (eat less well) and risk of heart failure (swollen leg). Methods This is a retrospective observational multicenter study. To gain insight on both short-and middle-term predictions and how the risk factors evolve through different periods of observation, we developed a series of models which predict the risk of future clinical symptoms.
This is a retrospective observational multicenter study. This study was conducted on two distinct cohorts. Data between January 2020 - February 2023 from 50 home care facilities using PRESAEGE CARE medical device on a daily basis were analyzed. 740 853 data from 27 439 visits by home aides for 1 478 patients. The patients' mean age was 84,89 years (SD = 8.9 years) with a moderate dependency level and the sample included 1 038 women (70%). PRESAGE CARE is a medical device CE marked to predict emergency hospitalizations. This e-health system is based on a questionnaire focused on functional and clinical autonomy (ie, activities of daily life), possible medical symptoms (eg, fatigue, falls, and pain), changes in behavior (eg, recognition and aggressiveness), and communication with the HA or their surroundings. Based on these data, some others risks are evaluated and predict by the artificial intelligence algorithm. This study aims to evaluate the sensitivity and specificity's predictions of PRESAGE CARE system for four clinical situations with a high impact on unscheduled hospitalization of olders adults living at home: falls, risk of depression (is sadder), risk if (eat less well) and risk of heart failure (swollen leg). The principal objective was the sensitivity and specificity of four events' prediction: falls, "is sadder", "eat less well" and "swollen leg" for non-tautological events (when events no appear in the observation window). Secondary objective was the sensitivity and specificity of four events' prediction: falls, "is sadder", "eat less well" and "swollen leg" for tautological events (when events appear in the observation window).
Study Type
OBSERVATIONAL
Enrollment
1,478
PRESAGE CARE is a medical device CE marked based on artificial intelligence to prevent and reduce emergency department visits and unplanned hospitalization among frail older adults living at home. These device is based on the use of a short questionnaire focused on functional and clinical autonomy (ie, activities of daily life), possible medical symptoms (eg, fatigue, falls, and pain), changes in behavior (eg, recognition and aggressiveness), and communication with the home care aides (HAs)or their surroundings. This questionnaire is composed of very simple and easy-to-understand questions, giving a global view of the person's condition. For each of the 27 questions, a yes/no answer was requested. Data recorded by HAs were sent in real time to a secure server to be analyzed by our machine learning algorithm, which predicted the risk level on emergency hospitalization risk and some health clinical situations and displayed it on a web-based secure medical device.
Sensitivity and specificity of four events' prediction: falls, "is sadder", "eat less well" and "swollen leg" for non-tautological events ((when events no appear in the observation window).
To evaluate the predictive performance of the models, we examined out-of-sample performance metrics, including area under the receiver operator characteristic curve (AUC), 5% and 15% AUC, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) based on confusion matric which was created as followed: At each point of the ROC curve we calculated weighted average between Sensitivity and Specificity.
Time frame: between one to six weeks
Sensitivity and specificity of four events' prediction: falls, "is sadder", "eat less well" and "swollen leg" for tautological events ((when events no appear in the observation window).
To evaluate the predictive performance of the models, we examined out-of-sample performance metrics, including area under the receiver operator characteristic curve (AUC), 5% and 15% AUC, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) based on confusion matric which was created as followed: At each point of the ROC curve we calculated weighted average between Sensitivity and Specificity.
Time frame: between one to six weeks
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.