New research from Harvard's Recovery Research Institute documents what precipitates relapse after years of sustained remission. The findings reshape what long-term recovery monitoring might actually look for.
For fifty years, relapse research has concentrated on the first months of recovery: stabilization, withdrawal, conditioned triggers, the world of "people, places, and things." That work has been essential. It also doesn't describe what happens to someone at year three, year seven, or year fifteen.
This research gap may have gotten smaller. In a new paper, “Long-Term Relapse: Markers, Mechanisms, and Implications for Disease Management in Alcohol Use Disorder” (Kelly et al., 2026, Frontiers in Public Health), John Kelly and colleagues at Mass General's Recovery Research Institute interviewed 50 adults who had relapsed on alcohol after being in remission for anywhere from one to twenty-three years. They asked what changed in the year before the relapse, and — crucially — how much each change contributed.
The answers were not consistent with early-recovery predictions in the literature.
What the research found
Across 50 participants, 26 distinct factors emerged as precipitants of long-term relapse, spread across four domains: biological, psychological, social, and recovery support services. Among the factors participants rated as definitely contributing to their relapse, the distribution looks like this:
Biology is the smallest slice. That is not what the conditioned-cue-reactivity model would predict, and it is the single most important thing the research is telling us: long-term return to use is overwhelmingly a psychological and social story.
The most commonly reported change, and the factor participants most often rated as definitely driving their relapse, was a decline in focus on recovery. Kelly's team describes it as a slow fading of cognitive recovery vigilance. The things that once sat at the center of daily life, such as meetings, rituals, check-ins, the sense of being a person in recovery; began to compete with new, often positive priorities: a promotion, a move, a relationship, a child. And then one day recovery wasn't the center anymore.
Mental-health symptoms, isolation, loneliness, and reduced mutual-help engagement rounded out the top risks. Life events — loss, job change, moves, financial stress — showed up in the middle of the ranking. Biological factors like sleep and energy were common but rarely rated as primary drivers; physical pain and recreational drug use were exceptions, rare but potent.
Precursors did not arrive suddenly. They accumulated and intensified across the year leading up to return to use. Most participants reported four or more definitely contributing factors spanning two or more domains.
Why this reframes long-term recovery monitoring
The clinical toolkit for early recovery — relapse-prevention coping skills, cue deconditioning, medication for withdrawal — was built for a different moment in the recovery arc. Kelly's paper argues, with evidence, that the drivers of long-term relapse are better explained by self-regulation theory, stress-and-coping theory, and social-identity theory than by neurobiological dysregulation. That reframe has real implications for what programs should measure, when, and how often.
It also surfaces a problem the field has been circling for a generation. The word recovery itself has come to mean so many different things to so many different people that it has become genuinely difficult to say, concretely, what is fading when someone's recovery is fading.
That is the frame William White offered the field years ago. But, this latest research sets the notion of "recovery" as an ongoing and singular construct inextricably tied to AUD in the disease model. Because of this, early recovery and even long-term recovery monitoring where it exists, seeks to suppress or prevent use. This research, albeit with a small sample, suggests we must measure and monitor the substrate variables – recovery capital – that sustain a person’s recovery.
What the field can act on today
Kelly et al. close with a call for "remission-based warning signs,” an empirically grounded checklist that clinicians and programs could use to monitor long-term recovery and intervene before symptomatic return-to-use. The factors they identified map cleanly onto the kinds of life-domain indicators that the Recovery Capital Index already captures: mental and emotional wellbeing, social support, isolation, purpose, community connectedness, safety, financial stability, employment, housing, healthy-lifestyle.
Here are three practical actions any program serving participants past the one-year remission mark can incorporate:
What we’re doing with this research
Commonly Well partners with organizations and communities in the use of the Recovery Capital Index (RCI) — a validated, participant-reported measure of the internal and external resources a person can draw from in recovery and growth. The RCI is in use at programs in behavioral health, recovery housing, treatment courts, and community organizations across the country, and it was explicitly built to operationalize William White's “antidote.”
Kelly et al. (2026) is, in a sense, a validation of that approach from the outside. Most of the 26 factors the study surfaces already live inside the RCI's structure, scored every time a participant completes an assessment. One finding: the drift in cognitive recovery vigilance; is a genuine gap in the current episodic care and support delivery structure. Through our provider and research partners we are actively working to extend the engagement and measurement lifecycle beyond early recovery.
We have prepared a companion practitioner guidance brief for our partners that maps Kelly's factors onto the RCI structure, identifies the gaps, and translates the findings into concrete monitoring actions for each of the program types we work with. If your team is thinking about how to act on this research, we'd welcome the conversation.
Want the practitioner guidance brief?
The companion brief covers the full Kelly-to-RCI crosswalk, archetype-specific practice guidance, a monitoring protocol with cadence and alert thresholds, and conversation prompts grounded in the research. It's available to partners and on request.
Email [email protected] or visit commonlywell.com
Primary source: Kelly, J. F., Klein, M., Zeng, K., Manske, S., & Abry, A. (2026). Long-term relapse: Markers, mechanisms, and implications for disease management in alcohol use disorder. Frontiers in Public Health, 13. doi: 10.3389/fpubh.2025.1706192
On the research: Cross-sectional, retrospective study of N=50. The authors characterize findings as preliminary and exploratory; prospective replication is needed. We share their caution and their conclusion that the signal is strong enough to inform practice now. A plain-language summary is also available at the Recovery Research Institute's Recovery Answers bulletin.
Work With Us
Commonly Well partners with behavioral health providers, community organizations, and health systems to conduct rigorous, actionable research that drives real outcomes.
Interested in similar strategy, analysis, and support for your organization?
Recovery Capital Index implementation and analysis
Community health surveys and population assessments
Program evaluation and outcomes measurement
Data dashboards for quality, performance, and equity monitoring
Contact Commonly Well | Schedule Call
