Design: retrospective cohort analysis of two US data sources with time-varying exposure Cox models
Population: 545 registry and 1,027 health-record patients with idiopathic pulmonary fibrosis starting antifibrotic therapy
Primary outcome: composite of 10% or greater FVC decline, hospitalisation, transplant or death
Effect: registry HR 1.17 (95% CI 0.90-1.52) for reduction and 1.51 (1.04-2.18) for discontinuation; health record 0.87 and 0.94

Two US sources - the Pulmonary Fibrosis Foundation registry and Duke University health system records - were used to follow 1,572 patients with idiopathic pulmonary fibrosis who started an antifibrotic, grouping them by whether they stayed on full dose, reduced for at least three months, or stopped for at least three months. Progression was a composite of a 10% or greater fall in per cent predicted forced vital capacity, hospitalisation, transplant or death.

About 60% in each source stayed on full dose. In the registry, progression rates per 100 person-years were 39.7 on full dose, 46.6 on reduced dose and 70.7 after stopping, with hazard ratios of 1.17 (95% CI 0.90-1.52) for reduction and 1.51 (1.04-2.18) for discontinuation, a set of comparisons that did not reach significance overall (p = 0.073). In the health record source the pattern disappeared entirely: 44.8, 40.5 and 41.7, with hazard ratios of 0.87 and 0.94.

The two sources disagreeing is the most honest finding here. If dose reduction carried a clear cost, both datasets should show it. The non-finding is useful in clinic, where the choice is not between full dose and reduced dose but between reduced dose and nothing - patients who cannot tolerate the full dose will stop altogether if the alternative is not offered. What both sources agree on is that progression was common whatever the dose, which is the point the authors make.

Offer a reduced antifibrotic dose rather than accepting discontinuation when tolerability is the problem.
About 40% of patients were not on full dose at follow-up in both sources.
The two data sources disagreed on whether dose reduction tracked with progression.
Progression was frequent in every dose group, which is the more robust finding.
This is observational: patients whose dose was reduced differ systematically from those it was not.

Why it matters

The real clinical choice is between a reduced dose and no drug at all, and this removes a reason to treat reduction as failure.

The statistics, in plain English

Two datasets pointing in opposite directions is a stronger caution than either result on its own, and it usually signals confounding that differs between the populations. The confounding here is obvious: clinicians reduce or stop the drug in patients who are struggling, so dose is partly a marker of how the patient is doing. Time-varying exposure modelling helps with that but does not remove it.