Mapping the path to lasting remission in lymphoma
A multi-state model of relapse and death gives real-world probabilities of durable remission after diffuse large B-cell lymphoma.
The published work
The challenge
For patients with diffuse large B-cell lymphoma (DLBCL), reaching a lasting remission — at least two years — is a strong signal of a favourable long-term prognosis. But patients and clinicians need real-world, interpretable probabilities of that chance, broken down by clinical characteristics, to support risk communication and decision-making.
A single survival curve can't capture the journey that matters here: remission, relapse, a possible second remission, and death. That calls for a model of the whole trajectory.
The approach
Red Door Analytics' Sara Ekberg and Michael Crowther followed 2,941 DLBCL patients who were in remission after primary treatment (Swedish Lymphoma Register, 2007–2014), using multi-state models to study patient trajectories through relapse and death, with flexible parametric models for the transition rates.
The analysis was implemented in Stata using RDA's own merlin and multistate packages — the very tools the team builds and maintains. It was published in British Journal of Cancer.
What it found
At two years, an estimated 80.7% of patients were predicted to remain in remission and 13.2% to have relapsed. Relapse risk peaked at around seven months, and the annual decline of patients in remission stabilised after two years.
Prognosis varied clearly by risk group: the probability of a lasting remission was 20.4 percentage points lower for patients with IPI 4–5 than IPI 0–1, with time in remission shortened by about 3.5 months. Patients who relapsed seldom achieved a durable second remission.
Why it matters
Multi-state models turn registry data into exactly the kind of interpretable, subgroup-specific probabilities a clinician can use in conversation with a patient — not just whether, but when and along which path.
It is also a real-world demonstration of RDA's software in the hands of its own authors: the flexible-parametric and multi-state machinery of merlin and multistate — the latter now succeeded by pendragon in R and Stata — applied to a question that mattered to patients.