Rainier Rehab Logo
Editorial illustration of a tall phone-shaped panel on a pale plane, thin concentric arcs…
November 15, 20264 min read

The Phone Can See a Relapse Coming. The Evidence Says It Cannot Say So Yet.

Methamphetamine has no approved medication with reliable effect, which leaves community treatment leaning on counseling, motivational work and contingency management — and on outpatient appointments spaced so far apart that the days before a relapse happen between them.

A narrative review published November 5 in Frontiers in Psychiatry by researchers at Kunming Medical University asked whether phones and wearables can close that gap. Assembled from 42 publication-level sources, the answer is that the sensors work, the participants mostly cooperate, and nobody has shown that the resulting alerts prevent anything.

Of the 42 core publications, only 13 involved cohorts formally diagnosed with methamphetamine use disorder, plus one drawing on a predominantly severe cohort. The rest came from broader substance use populations, mixed-cohort programs, or samples defined by past use rather than a diagnosis.

What the pilots measured

Ecological momentary assessment — the scheduled check-in asking about craving, mood, triggers and confidence — held up on feasibility. Outpatient cohorts completed roughly 65 percent of scheduled prompts across several weeks.

Then the results flatten. An eight-week pilot with counselor-reviewed assessments found no significant reduction in self-reported methamphetamine use or in urine toxicology. Across broader stimulant cohorts compliance decayed, and agreement between self-report and toxicology ranged from about 51 to 72 percent — a spread wide enough that self-report cannot carry a monitoring system alone.

The intervention trials behave the same way. An eight-week randomized trial of a smartphone app in 100 adults in community rehabilitation found no between-group difference in cue-induced craving or cognitive function, with 38.5 percent attrition by week 24. The S-Check trial randomized 259 people to immediate or delayed app access and found no difference in help-seeking or motivation at day 28. A six-month trial in 99 patients did report fewer positive urine screens with chatbot support than with usual care, but retention did not differ and abstinence after the trial was never assessed.

The most instructive study is the smallest. When researchers built machine-learning anomaly alerts from wearable and momentary data and handed them to peer recovery specialists, only 78 of 229 enrolled participants supplied at least 30 percent of their records; 63 reached the final comparison. The specialists also delivered identical check-ins for algorithm-triggered and randomly timed alerts while blinded to which was which, so the study could not show that risk-triggered outreach does anything a scheduled call does not.

Where the field has to go

The authors are more useful on design than on promise. They want population-level, individualized and hybrid risk models compared head to head on the same task instead of assuming personalization wins — hybrid models running population baselines while personal data is thin, then shifting to individual baselines recalibrated after a move or a sustained abstinence. They want precision-recall curves, calibration slopes and explicit warning lead times, since an alert arriving after the high-risk window is a data point, not a tool.

They are also unusually direct about governance. Passive sensing captures nocturnal movement, location and phone use — signals that overlap with the compliance criteria of probation, child welfare and employment, and participants who suspect the data feeds enforcement abandon the platform and evade the sensors. The standard they set is voluntary participation, data minimization, tiered consent, encrypted storage, and human review of every automated prompt.

For the stimulant side of the addiction crisis, where stimulant addiction treatment still rests on psychosocial care and contingency management carries the strongest randomized evidence, that is the honest position. Continuous monitoring may eventually tell a caseworker which week to call. It has not yet earned the right to tell anyone when to intervene.

RR
Rainier Rehab Editorial Team

Editorial Board

LADC, LCPC, CASAC

The Rainier Rehab editorial team consists of licensed addiction counselors, healthcare journalists, and recovery advocates dedicated to providing accurate, evidence-based information about substance abuse treatment and rehabilitation.

Related Articles