NOTE-1358: Lynch et al. — RefraMED trial of RO DBT for refractory depression

Contribution

It is the first multicentre randomised trial of radically open dialectical behaviour therapy, and the main controlled evidence the record holds for it. RO DBT, six months of weekly individual sessions and skills classes aimed at overcontrolled personality, added to usual care, beat usual care alone on depression at the end of therapy and did not at the pre-registered primary end-point five months later. The trial also reports gains on the two process measures that RO DBT targets, and an imbalance in serious adverse reactions that the authors put down to unequal reporting.

Key insight

The treatment works fast and the control catches up. Both arms improve throughout the 18 months; RO DBT’s improvement comes during therapy and then levels off, while usual care improves slowly at first and then faster. So the difference between arms is large at 7 months and shrinks after. What does not shrink is the gap on the measures of the targeted process: psychological inflexibility and emotional approach coping move apart between arms after therapy ends and stay apart. The trial can be read as showing that RO DBT changes what it says it changes, and that this change does not by itself hold the advantage on depression.

Assumptions

• Population. Adults with a current major depressive episode by SCID-I, an HRSD score of at least 15, and refractory depression: chronic (at least two years) or treatment-resistant (two or more previous episodes and no response to an adequate antidepressant dose for six weeks in the current one). Excluded: bipolar disorder, psychosis, “dramatic-erratic” (cluster B) personality disorders, primary substance dependence, current or awaited standard DBT. The exclusions are the undercontrolled presentations; overcontrol was not measured or required for entry. The trial is a trial of RO DBT in refractory depression, not in overcontrol.
• Setting. Three NHS secondary-care centres already running standard DBT (Dorset, Hampshire, North Wales); recruitment March 2012 to April 2015; 97% of responders White.
• Treatment as usual was whatever the NHS or private care offered except standard DBT. RO DBT participants were “strongly discouraged” from other psychotherapy during treatment; controls were not.
• Analysis. Three-level linear mixed models (time within patient within therapist), unbiased if data are missing at random; no multiple imputation. Bayesian refits with brms were done post hoc.

Key results

From Table 1 (adjusted means; difference RO DBT − TAU, so negative favours RO DBT on HRSD, AAQ-II, MSSI and PHQ-9):

• HRSD, primary. 7 months: 15.34 v. 20.73, −5.40 (95% CI −9.84 to −0.94), P = 0.023, effect size 1.03. 12 months (primary end-point): 14.19 v. 16.34, −2.15, P = 0.290, effect size 0.41. 18 months: 13.79 v. 15.48, −1.69 (−6.22 to 2.84), P = 0.424, effect size 0.32. The 12-month difference exceeds the two points the trial was powered to detect.
• AAQ-II (inflexibility). −3.37 (P = 0.096) at 7 months, −4.94 (P = 0.040) at 12, −5.48 (P = 0.014) at 18.
• Emotional approach coping. +1.50 (P = 0.132) at 7 months, +3.55 (P = 0.017) at 12, +2.98 (P = 0.012) at 18.
• PHQ-9. −3.45 (P = 0.041), −2.92 (P = 0.058), −3.23 (P = 0.030).
• Suicidal ideation (MSSI) and social support (SSQ-3). No significant difference at any point.
• Remission. Full remission was rare in both arms (RO DBT 1%, 8%, 7%; TAU 0%, 0%, 1% at 7, 12, 18 months). By a 17.5% fall in HRSD, 59%, 69%, 59% against 27%, 48%, 41%. Simulated for 100 new patients at 12 months: 32 improve by choosing RO DBT, 10 deteriorate, 58 unchanged.
• Power. Recruitment reached 250 of 276 (91%) and analysis 190 (76%) against 229 planned; the intratherapist correlation was 0.14 against the 0.025 assumed. The best and worst therapists differed by 2.6 HRSD points, more than the 12-month treatment effect.
• Post hoc Bayes. Posterior odds that RO DBT is better: 46 at 7 months, 5.5 at 12, 4.7 at 18.
• Harms. 28 serious adverse events in 24 RO DBT participants against 4 in controls. Eight, all in the RO DBT arm, were judged possibly (5) or probably (3) related, among them two overdoses and a prevented suicide attempt. No suicides. Therapists saw RO DBT participants twice a week and reported 23 of their 28 events; controls were seen only at assessments.

Claims

id	claim	strength	support
C1	RO DBT plus usual care reduces depressive symptoms more than usual care alone at the end of a six-month course	moderate	one trial; 7-month HRSD contrast, P = 0.023, 17 ratings made unmasked at that point
C2	The advantage on depression is not shown at 12 or 18 months	moderate	pre-registered primary contrast; underpowered by the authors’ own account
C3	RO DBT improves psychological flexibility and emotional coping, and the improvement persists to 18 months	moderate	AAQ-II and EAC, significant at 12 and 18 months; self-report; not significant at 7
C4	RO DBT is superior to usual care across the whole follow-up	weak	post hoc Bayesian posterior odds only
C5	The excess of serious adverse reactions in the RO DBT arm is a reporting artefact	weak	argued from unequal contact; not tested
C6	The trial’s title promises mechanisms; no mediation analysis is reported	strong	the paper itself: mediators were measured, only between-arm contrasts are given

Method

A pragmatic-explanatory hybrid trial, 3 : 2 allocation, stratified dynamic randomisation (onset age, HRSD above or below 25, personality disorder) by the Swansea Trials Unit, with participants also randomised among available therapists. Masked assessors; when a participant revealed the arm during an assessment the unmasked rating was kept (17 times at 7 months, 12 at 12). Twenty-three standard DBT therapists were trained for ten days by the treatment’s developer, who supervised them and did not treat. Fidelity was rated on the standard DBT adherence scale, there being no RO DBT one: 221 of 273 rated tapes (81%) adherent. RO DBT participants attended on average 22.9 of 29 individual sessions and 19.9 of 27 classes.

Concepts

• Overcontrol — “excessive inhibitory control”: cognitive and behavioural rigidity, a strong desire to control one’s environment, restrained emotional expression, limited social interaction, and aloofness, mistrust and fear of criticism in close relationships. The paper places it in cluster A and C personality disorders.
• RO DBT — a “transdiagnostic” therapy that, unlike other therapies, encourages “social bonding through emotional expression”; the paper’s strengths section calls it “the first treatment known to us to target deficits in social signalling as the main problem underlying overcontrolled emotional loneliness”.
• Refractory depression — chronic or treatment-resistant, as above.

Connections

The treatment is Lynch’s, from his 2018 manual (reference 9, not registered), with standard DBT skills lessons from Linehan’s 1993 skills manual (reference 17, not registered); the record’s DOI’d statements of the two are LIT-1205 and LIT-1221. Two earlier pilots by Lynch with older depressed adults (2003, 2007) used an earlier form of the treatment and gave the effect sizes (0.85, 0.71) the power calculation assumed. The systematic review that places this trial among the RO DBT evidence is LIT-1224. Its measure of inflexibility, the AAQ-II, is the acceptance and commitment therapy measure, so the trial’s process gain is stated in ACT’s terms.

Bearing on the record

• CLAIM-349. The claim cites A42 for RO DBT’s evidence being “promising but not definitive” and RefraMED’s longer-term findings being “less conclusive”. The reading confirms both. Two things in it bear on the claim more specifically. First, the trial selected on refractory depression with the undercontrolled presentations excluded; it did not measure overcontrol, so it is no evidence that overcontrol is common in refractory depression or that the treatment’s effect runs through it. Second, the change that persisted was on flexibility and emotional expression, the two properties CLAIM-349 makes the mark of adaptive regulation, while the change on depression did not. That is consistent with the claim’s own caution, in “What it does not say”, that flexibility is not the whole of the outcome, and it does not test the claim, which is about organizations. The LIT’s “Why this is here” reads the trial as “a caution on reading overcontrol as the whole of the problem”; that is the record’s reading, not the paper’s, which concludes the opposite way, that RO DBT’s targets are the right ones.
• QUESTION-045 asks whether protective lock-in is the same mechanism at the individual scale. The trial gives nothing on mechanism: the mediators were collected and no mediation is reported.

No instruction for machine-learning practice; nothing for the anthology.

Limitations

• Underpowered at the end-point it chose, by the authors’ account: short recruitment, more loss to follow-up than planned, and therapist heterogeneity nearly six times the assumed value.
• Unequal co-treatment. RO DBT participants were discouraged from other psychotherapy; controls could seek any treatment except standard DBT. Usual care’s late improvement may be partly that.
• Masking. Seventeen ratings at 7 months, the time point where the effect was found, were made by assessors who knew the arm (12 at 12 months, none at 18). The paper does not say how they split between arms.
• Developer and financial interest. Six of sixteen authors have received royalties or fees for RO DBT; the developer trained and supervised the therapists and leads the paper; a co-author co-owns the training company.
• Fidelity was measured with the standard DBT scale, not one for RO DBT.
• Process measures are self-report, from participants who knew their arm, and the processes are the ones the therapy teaches.
• Errors and slips found.
  • “Throughout RO DBT participants reported significantly better psychological flexibility and emotional coping than controls” (the abstract; the Results and Discussion say the same). Table 1 has the 7-month differences at P = 0.096 (AAQ-II) and P = 0.132 (EAC): the gain is significant at 12 and 18 months, not throughout.
  • Table 1 gives the 12-month HRSD interval as −6.69 to 2.28; the abstract and text give −2.28 to 6.59. The point estimate −2.15 sits at the centre of −6.59 to 2.29, so the table’s 6.69 is a typo.
  • The one-tailed Fisher test for eight serious adverse reactions in 162 participants against none in 88 is given as P = 0.004. With eight participants affected, P is C(162,8)/C(250,8) = 0.029; with fewer it is larger. 0.004 is not reproduced from the numbers given. It is still under 0.05.
  • References 5 and 8 are the same paper (Riso et al. 2003), and the second is the one cited for the core characteristics of overcontrol, which that paper, on cognitive aspects of chronic depression, is not obviously the source of.
  • The title’s “mechanisms” are not reported: no mediation analysis is in the paper.

Open questions

• Does the effect on depression run through the change in flexibility and emotional coping? The data (Zenodo record 1442883) would allow a mediation analysis the paper does not give.
• Does RO DBT do better in participants selected for overcontrol than in refractory depression at large? The trial did not measure it.
• With equal permission to seek other therapy and an RO DBT fidelity scale, does the 12-month difference hold at the two points the trial sought?

Note of 2026-10-29: the monograph, read

The NIHR monograph, reference 13, which this note’s history lists as not read, is now read: LIT-1838, NOTE-1421. It bears on three findings here.

• “Throughout.” The monograph’s Table 8 supports “throughout” (p = 0.010 and 0.036 at 7 months), with the same differences and narrower intervals. So the abstract’s “throughout” follows the monograph’s analysis of the same data, while this paper’s Table 1 is the more conservative analysis. The mismatch within this paper stands, and the abstract should have matched its own table.
• P = 0.004. The monograph gives “p < 0.01”. Neither reproduces the 0.029 computed here, and neither says how many participants had the eight reactions.
• The first open question, whether the effect on depression runs through flexibility and emotional coping, is not answered. The monograph’s mediators are sessions attended, alliance and DBT skills, not AAQ-II or EAC, and its instrumented estimates failed on weak instruments. The cross-lagged analyses it planned are not reported.