New on CRAN – mlumr: Bayesian multilevel unanchored indirect treatment comparisons in R In health technology assessment (HTA), many times there is no common comparator between two treatments of interest. This is fairly common in oncology, rare diseases, and accelerated approvals in which many trials are single-arm. Matching-adjusted treatment comparison (MAIC) and simulated treatment comparison (STC) are common techniques for unanchored situations where each has trade-offs. Multilevel unanchored meta-regression (ML-UMR) was introduced by Conor Chandler and Jack Ishak at ISPOR Europe 2025. ML-UMR extends multilevel network meta-regression (ML-NMR), developed by Phillippo et al., to unanchored settings. ML-UMR adjusts populations by integrating an IPD outcome model over the comparator’s covariate distribution (using Sobol sequence quasi-Monte Carlo (QMC) Gaussian copula). This way, both IPD and aggregated data can be modeled in the same regression model. Now, we are proud to announce that mlumr R package has been released. This package comes with: ✅ Two model variants: Shared prognostic factor assumption (SPFA) and relaxed (treatment-specific prognostic factors). ✅ Three outcome families: Binary (logit/probit/cloglog), continuous (identity/log), count (log). ✅ Marginal and conditional effects in either index or comparator population. ✅ Model comparison (LOO/WAIC/DIC) and prior_sensitivity() for relaxed models. ✅ STC (using G-computation) and naïve benchmarks. ✅ Two engines: RStan by default, optional CmdStanR backend. It’s free software, licensed under GPL-3, and on CRAN today: install.packages(“mlumr”)

Documentations: choxos.github.io/mlumr Code & issues: github.com/choxos/mlumr

Transparency is the new objectivity. David Weinberger (in a 2009 blog post, cited in Andrew Gelman and Christian Hennig's paper "Beyond subjective and objective in statistics") argues that objectivity functioned as a trust mechanism for a medium that couldn't easily show its workings. If a journalist claimed to be "objective," readers often had little choice but to accept it, as following sources or assumptions was costly and time-consuming. In that setting, objectivity served as a kind of stopping point: you had reached a credible source, and the inquiry could reasonably end. That model made sense for the paper. It’s less clear that it works as well now. Gelman and Hennig make a related point for statistics. The labels "objective" and "subjective" don't do as much useful work as we might hope. In practice, they often obscure the real issues: what assumptions were made, how sensitive the results are, and how alternative choices might change conclusions. Instead, they suggest focusing on more concrete qualities: transparency, impartiality, stability, awareness of multiple perspectives, and alignment with observable reality. There's also empirical support for why this matters. In a large "many analysts" study, Rotem Botvinik-Nezer and colleagues (2020) gave the same dataset to dozens of independent teams. No two teams used identical pipelines, and their conclusions often differed (sometimes substantially) despite analyzing the same data. That kind of variability is hard to reconcile with the idea that any single analysis is simply "objective." Seen this way, some familiar practices look a bit different: - A "default" prior (in Bayesian statistics) isn’t neutral; it's a choice that can be explained (or not). - A p-value depends on modeling decisions, stopping rules, and reference sets. - Even frameworks designed to minimize judgment still embed assumptions. None of this invalidates these tools, but it does suggest that calling them "objective" may not be the most informative description. What seems more useful, at least in many cases, is making those choices visible: - What assumptions were made? - What alternatives were considered? - How sensitive are the results? - What would change under different reasonable decisions? Transparency doesn't remove bias or disagreement. But it gives others a way to understand, critique, and, if needed, re-run the reasoning. So perhaps "transparency is the new objectivity" is less a replacement than a change in emphasis: Not "trust this because it’s objective," but "here’s how this was done; see for yourself!" It’s a more demanding standard, and probably an ongoing one rather than a box to check.

It's been more than two weeks since the start of the protests in Iran and more than 3 days since complete internet blockout. Reports confirm murdering of thousands of Iranians which is a public health crisis, yet the mainstream medical journals are in complete silence. In comparison to Gaza which has been the main topic of @TheLancet for a long time (1,027 articles to date, with the EiC as the main contributor with 40 articles), there is no single article on Iran's situation as of this moment. Two years ago, I shared a post on my blog about this bias among top medical journals indicating some lives are more equal than the others. choxos.com/research-top-medi…