There’s a reason they’re called smartphones.
Indeed, how patients use their smartphones and where they take them provides insight into what has been termed their “digital phenotype.” It’s information that, analyzed correctly, becomes useful in differentiating bipolar disorder from borderline personality disorder, a distinction that’s often challenging in clinical practice, Kate E.A. Saunders, MD, DPhil, said at the virtual congress of the European College of Neuropsychopharmacology.
Dr. Saunders, a psychiatrist at the University of Oxford (England), and colleagues have developed a smartphone app enabling patients to briefly characterize their current mood on a daily basis, as well as a machine learning model to analyze this data stream as patients’ moods evolve over time. In their prospective longitudinal Automated Monitoring of Symptom Severity (AMoSS) study of 48 patients with a confirmed diagnosis of bipolar disorder, 31 with borderline personality disorder, and 51 healthy volunteers, the tool correctly classified 75% of participants into the correct diagnostic category on the basis of 20 daily mood ratings (Transl Psychiatry. 2018 Dec 13;81:274. doi: 10.1038/s41398-018-0334-0).
The app also monitors activity via accelerometry and geolocation to assess an individual’s circadian rest-activity patterns, as well as telephone use and texting behavior. In another report from AMoSS, Dr. Saunders and coinvestigators showed that these patterns also distinguish persons with bipolar disorder from those with borderline personality disorder, who in turn differ from healthy controls (Transl Psychiatry. 2019 Aug 20;91:195. doi: 10.1038/s41398-019-0526-2).
It doesn’t replace doctors, but clearly it can add to diagnostic accuracy,” she said.
Borderline personality disorder and bipolar disorder are common diagnoses with quite different treatment approaches and prognoses. Studies have shown that rates of misdiagnosis of the two disorders are significant. The challenge is that they share overlapping diagnostic criteria, including prominent mood instability, which is difficult to assess reliably in clinical practice. That’s because the assessment relies on retrospective self-report of how patients felt in the past, which is often colored by their present mood state. The smartphone app sidesteps that limitation by having patients rate their mood daily digitally across six categories – anxiety, elation, sadness, anger, irritability, and energy – on a 1-7 scale.
The machine learning model that analyzes this information organizes the voluminous data into what Dr. Saunders called “signatures of mood” and breaks them down using rough path theory, a mathematical concept based upon differential equations. Dr. Saunders and colleagues have demonstrated that the shifting daily mood self-rating patterns can be used not only to sharpen the differential diagnosis between bipolar disorder and borderline personality disorder, but also to predict future mood. Automated analysis of the past 20 previous mood self-ratings predicted the next day’s mood in healthy controls with 89%-98% accuracy, depending upon which of the six mood categories was under scrutiny.
The predictive power in patients with bipolar disorder was also good, ranging from 82% accuracy for the energetic and anxious domains to 90% for the angry mood category. This ability to predict future mood states could have clinical value by assisting bipolar patients in enhancing proactive self-management and managing their mood stability to avoid depressive or manic relapse, although this has yet to be studied.
“For borderline personality disorder the predictive accuracy was not so good – 70%-78% – but perhaps that doesn’t matter,” Dr. Saunders said. “Perhaps that difficulty in predicting mood may actually be quite a useful diagnostic marker.”