Design: multicentre, double-blind, sham-controlled randomised clinical trial over 12 months
Population: 93 adults with major depressive disorder or bipolar disorder (80 in the modified intention-to-treat analysis)
Primary outcome: number of recurrent mood episodes per participant during follow-up
Effect: incidence rate ratio 3.39 favouring the active app (95% CI 1.86 to 6.17); hazard ratio for time to recurrence 3.03 (95% CI 1.58 to 5.81)

Ninety-three adults with major depressive disorder or bipolar disorder were randomised, double-blind, to an active circadian rhythm stabilisation app or a sham app with an identical interface, and followed for twelve months. The active app used passive smartphone sensor data to generate individualised three-day mood forecasts and machine-learning feedback on circadian behaviour; the sham produced non-actionable feedback from a dummy algorithm. The primary outcome was the number of recurrent mood episodes per participant.

In the modified intention-to-treat sample of 80, recurrence was substantially higher on the sham app: incidence rate ratio 3.39 (95% CI 1.86 to 6.17). Cumulative recurrent episode-days per person-year were also higher on sham, duration rate ratio 2.76 (95% CI 1.19 to 6.40), and time to recurrence favoured the active app, hazard ratio 3.03 (95% CI 1.58 to 5.81). No significant adverse effects were reported.

This is an unusually clean digital trial. Sham-controlled, double-blind, a visually identical comparator, and a hard clinical outcome over a full year rather than a symptom scale at eight weeks — which is what most app studies offer. The effect size is larger than most pharmacological maintenance data, which is itself a reason for caution.

The transferable part is not this specific app, which is not commercially available in India. It is the target: circadian regularity as a maintenance intervention, measured and fed back rather than advised once at discharge. Sleep-wake timing, light exposure and meal timing are already within reach of every follow-up appointment, and this trial is the strongest argument yet for spending clinic minutes on them.

Make sleep-wake timing, not just sleep duration, a standing item at maintenance follow-up in mood disorders
Ask for the actual bedtime and wake time over the last week rather than a usual pattern
Set a regular wake time as the anchor — it is easier to hold than bedtime and drives the rest
Use any passive step or sleep data the patient already collects, rather than waiting for a validated app
Treat this as an adjunct: every participant remained on standard care

Don't overread it

One trial of 93 participants at a single research network — this shows a specific app worked, not that any commercial sleep app will.

The statistics, in plain English

An incidence rate ratio of 3.39 means recurrences occurred over three times as often in the sham group, and the interval, 1.86 to 6.17, is wide — which is what 80 participants and a count outcome give you. The direction is secure; the magnitude is not, and a threefold difference from an app should be treated as an upper estimate until replicated. Thirteen of 93 randomised participants are missing from the modified intention-to-treat analysis, which matters more in a recurrence count than it would in a symptom score.