This article is more than 5 years old. Information may no longer be current.
Smartphone app, algorithm offer insights into symptom burden among chemotherapy-treated patients
Click Here to Manage Email Alerts
Patients with cancer who undergo chemotherapy can be monitored remotely with smartphone sensors and an algorithm that detects worsening symptoms based upon objective changes in patient behavior, according to results of a pilot study.
“On days when patients reported worse-than-average symptoms, they tended to spend more time being sedentary, moved the phone more slowly and spent more minutes using apps on the phone,” Carissa A. Low, PhD, assistant professor of medicine and psychology at University of Pittsburgh, said in a press release. “Collecting these objective behavioral measures from smartphone sensors requires no additional effort from patients, and they could prove beneficial for long-term monitoring of those undergoing arduous cancer treatments or those with other chronic illnesses.”
Real-time monitoring and management of common physical and psychological symptoms among patients with cancer who receive chemotherapy could lead to improvements in overall patient outcomes.
Low and colleagues assessed whether data collected by a smartphone app and Fitbit activity monitor could be used to estimate daily symptom burden among 14 patients undergoing chemotherapy for gastrointestinal cancer.
Researchers asked patients to carry a smartphone for 4 weeks. A smartphone app passively and continuously collected data on patient behavioral patterns, including the number of calls or texts sent and received, smartphone apps used, and the movement and location of the phone.
Investigators then averaged patient-reported symptom severity ratings to create an overall daily symptom burden score categorized into low, average or high symptom burden days. They pooled data from the smartphone to develop an algorithm that could identify and correlate symptom days.
The algorithm resulted in an overall accuracy of 88%. Behavioral features that best predicted high patient-reported symptom burden days included sedentary behavior as the most common activity, fewer minutes of light physical activity, less variable and average acceleration of the phone, and longer screen-on time and interactions with apps on the phone.
Smartphone features appeared associated with better predictive ability than Fitbit features, according to the researchers.
“These findings highlight opportunities for long-term monitoring of [patients with cancer] during chemotherapy, with minimal patient burden as well as real-time adaptive interventions aimed at early management of worsening or severe symptoms,” Low and colleagues wrote.
HemOnc Today spoke with Low about the study, how the data generated by the app can be used to monitor patients and potentially allow for earlier intervention if problems arise, and what still needs to be addressed in subsequent research.
Question: How did you conduct the study?
Answer: We were interested to find out whether we could detect fluctuations in how patients feel during chemotherapy based upon changes in their behavior — for example, how much they move around or how many phone calls they make in any given day.
For this initial pilot study, we recruited patients who were undergoing chemotherapy for upper or lower gastrointestinal cancer. We provided them with a Fitbit device, as well as an Android smartphone unless they already owned one. We used an app installed on the smartphone to securely, continuously and passively collect information about the movement of the patient and patient disease patterns. We used the Fitbit to monitor step counts and sleep for 4 weeks, which represented two chemotherapy cycles. One to two times per day, the app would ping patients and ask them to rate the severity of 12 common symptoms, including nausea, fatigue and pain. We then examined whether passively collected data about mobility and activity levels, communication patterns, phone use and sleep co-varied with changes in patient-reported symptoms, and whether we could use the passively collected data alone to classify higher-than-average or lower-than-average symptom burden day for each patient.
Q: What did you find?
A: When we pooled data from all 14 patients, we found that our classification model was 88% accurate for differentiating high from average or low symptom burden days. This means we were able to accurately detect fluctuations in how patients reported that they felt based solely on information that we could infer about their movement and their phone use behavior patterns. On days when patients had more fatigue, nausea or other symptoms than what was typical for them, they spent more time being sedentary, they moved the phone around more slowly, and they spent more time interacting with apps on the phone.
Q: How can this information be used to monitor patients and potentially intervene earlier when problems arise?
A: We ran our analysis after data collection was complete, but it would be possible to run the classification in real time as data from the app and Fitbit are collected. In this case, if high symptoms are detected, the app can alert providers to let them know that the patient is experiencing high levels of toxicities. It also can be used to message patients with suggestions for symptom management strategies or suggest they contact a specific provider. Because the algorithm relies on data that can be collected passively without any extra effort or attention from patients, it can be possible to remotely monitor patients for a long period of time without further burdening them at a time when they are already feeling overwhelmed. It also may be able to detect symptoms and fluctuations before a patient feels so sick that they wind up in the ER, or before their next scheduled oncologist appointment, so worsening symptoms can be managed earlier and hopefully more effectively.
Q: What still needs to be addressed in future research?
A: This study was small and preliminary. We plan to follow up on these results with a larger sample of patients and for a longer period of time to see if our classification model generalizes. It would be interesting to look at other outpatient treatments, as well, such as immunotherapies. It also will be important to investigate how feasible and acceptable real-time monitoring might be, and whether using the predictions that our model generates to trigger alerts to patients and providers actually can reduce patient symptoms or stress levels during chemotherapy. To do this, we will need to work collaboratively with both patients and providers to ensure that any system that we develop meets their needs and fits into the fabric of their busy daily lives.
Q: Is there anything else that you would like to mention ?
A: There is a lot of potential for leveraging data with smartphones. These devices are capable of collecting these data, which can tell us a lot about changes in behavior that may be clinically meaningful. The big challenge is trying to figure out how to integrate this patient-generated data into clinical oncology care. – by Jennifer Southall
Reference:
Low CA, et al. J Med Internet Res. 2017;doi:10.2196/jmir.9046.
For more information:
Carissa A. Low, PhD, can be reached at University of Pittsburgh, Suite 614, UPMC Hillman Cancer Center, 5200 Centre Ave., Pittsburgh, PA 15232; email: lowca@upmc.edu.
Disclosure: Low reports no relevant financial disclosures.