Changes in sleep can signal hypomanic episodes days before they begin, study finds

A new study published in the Journal of Affective Disorders has found that subtle fluctuations in sleep and activity levels may anticipate the onset of hypomanic episodes in people with bipolar disorder. By using wearable devices and self-report tools over the course of a year, researchers were able to detect early warning signs of mood changes up to three days before a full episode was reported.

Bipolar disorder is a psychiatric condition marked by shifts between periods of depression and periods of elevated or irritable mood, known as mania or hypomania. While depressive episodes often receive the most clinical attention, hypomanic episodes—periods of unusually high energy, reduced need for sleep, and impulsive behavior—can be disruptive and increase the risk of full-blown mania. Detecting these mood changes before they escalate could support early intervention strategies and improve outcomes.

The researchers conducted the study because past attempts to predict mood episodes in bipolar disorder have shown mixed results. Some studies have suggested that changes in sleep and activity precede mood shifts, while others have found inconsistent results when comparing clinician ratings to data from smartphones or wearable devices. A major obstacle has been the inability to track behavior frequently and precisely enough to identify early signs. This study sought to overcome that gap by using continuous monitoring and advanced analytic techniques.

The study followed 164 individuals diagnosed with bipolar I or bipolar II disorder, although only 50 experienced at least one hypomanic episode and were included in the final analysis. Participants were recruited from outpatient programs at two academic medical centers in Canada. All participants were adults diagnosed using standard clinical interviews and were free from active substance use disorders. After enrollment, participants were monitored for over a year, with some extending participation to two years.

Each participant wore an Oura ring—a commercial wearable device that continuously records physiological data, including movement and sleep. The device provided detailed, high-frequency information, capturing 5-minute intervals of energy expenditure and sleep stages. In addition to the passive data collection, participants completed weekly online self-reports using the Altman Self-Rating Mania Scale, which assesses five key symptoms of hypomania: elated mood, inflated self-confidence, reduced need for sleep, talkativeness, and heightened activity levels. Participants also completed the Patient Health Questionnaire to monitor depressive symptoms.

The researchers defined a hypomanic episode as any one-week period in which participants reported experiencing three or more hypomanic symptoms most days, based on a total scale score of 10 or more. Using these self-reports as a benchmark, the team then analyzed wearable data to see whether patterns in sleep and activity changed in the days leading up to an episode.

To do this, the researchers used a data analysis technique that breaks down complex time series into simpler components. By measuring how much these patterns fluctuated from day to day or within a single day, the team aimed to identify “spikes” in variability that might signal an impending mood shift. They defined a correct prediction as any significant change in sleep or activity that occurred within seven days before the onset of a hypomanic episode.

The analysis revealed that both sleep and activity levels began to show unusual variability a few days before participants reported symptoms of hypomania. On average, day-to-day changes in sleep were detected about three days before a reported episode, while changes in activity showed up about two and a half days in advance. Within-day fluctuations (such as 12-hour patterns) were particularly effective at identifying early shifts, outperforming weekly summaries or average values.

When comparing the effectiveness of different signals, 12-hour variability in sleep achieved a balanced accuracy of 87%, while activity data reached 89%. Sensitivity—the ability to correctly identify true positive cases—was above 90% in both cases, indicating that these methods detected many of the oncoming episodes. The results were even stronger when specific symptoms were analyzed. For instance, changes in sleep most accurately predicted the “decreased need for sleep” symptom, while changes in activity closely matched the “increased activity” item on the mania scale.

Interestingly, participants who experienced hypomanic episodes tended to be younger than those who did not, but there were no significant differences in gender, education, marital status, or baseline mood symptoms. The group was diverse in terms of gender identity and socioeconomic background, and most participants were receiving some form of psychiatric treatment during the study.

The researchers noted that variability—rather than average levels—was the most informative signal. This finding supports newer approaches to mental health that view mood as a dynamic system, where stability is maintained until disrupted by a loss of balance. In this model, rapid shifts between states, such as from stable mood to hypomania, may occur when a person’s usual regulatory mechanisms are no longer able to adapt. Detecting those disruptions in real time could be key to preventing more severe outcomes.

The study’s strengths include the use of continuous, high-resolution data from a wearable device, as well as a large number of data points over a long observation period. Unlike many earlier studies, this one did not rely solely on retrospective self-reports or clinician judgments, making the findings more reliable. The fact that the entire study was conducted remotely also highlights the feasibility of using wearable technology in real-world clinical settings.

However, the researchers acknowledged some limitations. The sample size, while robust for this type of intensive monitoring, was still relatively small compared to broader population studies. In addition, the study did not control for demographic factors like age or socioeconomic status, which could influence behavior. The team also noted that wearable devices can sometimes have missing or inconsistent data, which they addressed using advanced statistical techniques.

Future research could build on this study by including larger, more diverse samples and examining whether similar patterns hold for depressive episodes or mixed states. It may also be useful to explore how environmental factors, such as stress or social interactions, interact with biological signals to predict mood changes.

The study, “Day-to-day variability in sleep and activity predict the onset of a hypomanic episode in patients with bipolar disorder,” was authored by Abigail Ortiz, Ramzi Halabi, Martin Alda, Almendra Burgos, Alexandra DeShaw, Christina Gonzalez-Torres, Muhammad I. Husain, Claire O’Donovan, Mirkamal Tolend, Arend Hintze, and Benoit H. Mulsant.