End Stage Cancer, Terminal Cancer
Conditions
Keywords
palliative, cancer, survival prediction, artificial intelligence, wearable devices
Brief summary
This study is going to use wearable devices and smartphones to collect physical data from terminal patients and build a survival predicting model for terminal patients with machine learning. Investigators hypothesize that continuous physical data monitoring could offer a hint to better predictability in end-of-life care.
Detailed description
The study aim to examine the feasibility of utilizing wearable devices and smartphones in palliative patients in Taiwan. In addition, investigators try to identify the relationship between mobile health data and disease progression and establish a predicting model to the emergent medical need and death of patients, via machine learning. This is a single-arm observational study using wearable devices and smartphones in terminal cancer patients. Investigators planned to enroll 75 patients who receive palliative care. After obtaining consent from the patients or their legally authorized surrogate decision-makers, a baseline assessment will be conducted, with a guide to use wearable devices and phone apps. Investigators will keep regular follow-up for 52 weeks or until the participants' death. Assessment will be conducted every week, face-to-face or by telephone contact. A routine assessment includes symptoms and functionality in the past week, and vital signs and facial photograph will be recorded if possible. Physical data measured from wearable devices would be recorded continuously. The emergent medical needs of patient, including emergency department visit, unplanned admission and death of participants will be recorded if happen. The primary outcome is the predictive performance (sensitivity and specificity) of the machine-learning model using wearable device data and symptoms assessment. The secondary outcomes are symptoms, including pain, dyspnea, diarrhea, constipation, nausea, vomiting, insomnia, depression, anxiety and fatigue. Users' opinion and comment to using experience will also be recorded.
Interventions
None listed
Sponsors
National Taiwan University
National Taiwan University Hospital
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
20 Years to 105 Years
Healthy volunteers
No
Inclusion criteria
* Age: 20 years old or older * Clinical diagnosis: cancer in terminal stage.
Exclusion criteria
\- Cannot cooperate with use of wearable devices or smartphones.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Area Under the Receiver Operating Characteristic curve (AUC-ROC) of the machine-learning model to predict survival using wearable device parameters and clinical assessment | From date of enrollment until the date of death, or assessed up to 26 weeks. Wearable device parameters are collected continuously. Other clinical assessments are performed every week. Death or survival is recorded at the time the case closed. | Measured data from wearable device and regular assessment (including medical condition, laboratory data, symptom, functional assessment) will be integrated to build one machine-learning model to predict patients' death or survival within specific time range. The primary outcome is to evaluate the Area Under the Receiver Operating Characteristic curve (AUC - ROC) of the machine-learning model in predicting patients' survival. |
| Area Under the Receiver Operating Characteristic curve (AUC-ROC) of machine-learning model to predict unexpected medical needs using wearable device parameters and clinical assessment | From date of enrollment until the date of death, or assessed up to 26 weeks. Wearable device parameters are collected continuously. Other clinical assessments are performed every week. Events are recorded upon happening or afterwards. | Measured data from wearable device and regular assessment (including medical condition, laboratory data, symptom, functional assessment) will be integrated to build one machine-learning model to predict patient's unexpected medical needs (which is defined as emergency department visit or unplanned admission to hospital). The primary outcome is to evaluate Area Under the Receiver Operating Characteristic curve (AUC-ROC) of the machine-learning model in predicting unexpected medical needs. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Correlation between symptoms and wearable device parameters | From date of enrollment until the date of death, or assessed up to 26 weeks. Wearable device parameters are collected continuously. Symptoms assessed every week. | The severity of symptoms will be recorded by symptoms assessment scale (SAS). Investigators will explore the correlation between the wearable device parameters and symptoms. |
| Correlation between Australia-modified Karnofsky Performance Status (AKPS) and wearable device parameters | From date of enrollment until the date of death, or assessed up to 26 weeks. Wearable device parameters are collected continuously. Functional status assessed every week. | The functional status will be assessed by Australia-modified Karnofsky Performance Status (AKPS) during the follow-up. Investigators will explore the correlation between AKPS and wearable device parameters |
| Correlation between palliative care phase and wearable device parameters | From date of enrollment until the date of death, or assessed up to 26 weeks. Wearable device parameters are collected continuously. Palliative care phase assessed every week. | Evaluation of palliative care phases from the Palliative Care Outcomes Collaboration (PCOC) system will be assessed regularly. Investigators will explore the correlation between the palliative care phases and other parameters (wearable device parameters, symptoms, medical condition). |
Other
| Measure | Time frame | Description |
|---|---|---|
| Duration between events | From date of enrollment until the date of death, or assessed up to 26 weeks. Duration was calculated after cases closed. | Investigators will record duration between events (death, unexpected medical needs, admission and discharge) or duration from enrollment to events, if they happen |
| Overall survival and survival time | From date of enrollment until the date of death, or assessed up to 26 weeks. Calculated after all cases closed. | Investigators will record the overall survival and survival time from enrollment. |
| Comparison of AUC-ROC in survival prediction between machine learning model and Palliative Performance Scale (PPS) | From date of enrollment until the date of death, or assessed up to 26 weeks. PPS are assessed every week. Death or survival is recorded at the time the case closed. | Palliative performance scale (PPS) will be regularly assessed during the follow-up. The AUC-ROC of using PPS for survival prediction will be calculated and compared with the machine-learning model. |
| Tolerability and user experience to wearable devices | Assessed at the time the case closed | Investigator will ask and record any discomfort or side effect noted during the follow-up and at the end of the study. Investigator will survey for user experience of patients or caregivers at the end of the study. |
| Relation between personal background and user experience of wearable devices | Assessed at the time the case closed | Personal background such as educational level, age, and previous use of technological product will be recorded. Investigator will explore the relation between these factors and the user experience. |
| Site of death | Assessed at the time the case closed, only if the patient died | If patient died during the follow-up, investigator will record the site of death (at home or any other chosen place, in the hospital or ER). Other details will be recorded if the family or caregivers are willing to provide. |
| Comparison of AUC-ROC in survival prediction between machine learning model and Glasgow Prognostic Score (GPS) | GPS assessed retrospectively if data available. Death or survival is recorded at the time the case closed. | Glasgow Prognostic Score (GPS) will be assessed if C-reactive protein (CRP) and albumin are examined during the follow-up. The AUC-ROC using GPS for survival prediction will be calculated and compared with the machine-learning model. |
| Comparison of AUC-ROC in survival prediction between machine learning model and Palliative Prognostic Index (PPI) | From date of enrollment until the date of death, or assessed up to 26 weeks. PPI are assessed every week. Death or survival is recorded at the time the case closed. | Palliative Prognostic Index(PPI) will be regularly assessed during the follow-up. The AUC-ROC of using PPI for survival prediction will be calculated and compared with the machine-learning model. |
| Comparison of AUC-ROC in survival prediction between machine learning model and Palliative Prognostic Score (PaP) | From date of enrollment until the date of death, or assessed up to 26 weeks. PaP assessed every week only if the laboratory data available. | Palliative Prognostic Score (PaP) will be assessed if laboratory data available during the follow-up. The AUC-ROC of using PaP for survival prediction will be compared with the machine-learning model. |
| Time spent at medical service | Recorded when events happen or afterwards | If unexpected medical needs happen, investigators will record time spent at ER stay or hospital admission |
Countries
Taiwan
Contacts
Primary ContactJen-Hsuan Liu, MD
Backup ContactJaw-Shiun Tsai, MDPHD
Outcome results
None listed