Scientific Programme
Joint DSTS / DSBS Meeting 2026
- Dates 19th - 20th November, 2026
- Venue
Novo Nordisk Gentofte
Niels Steensensvej 2G, Gentofte - Registration CLOSED
Day 1
Day 1 · Thursday, 19th Novemeber Observational studies · Randomized studies · AI
11:00 – 12:00
Arrival and registration
12:00 – 12:15
Welcome and opening remarks
Session 1 — Observational Studies
Session 1 · 12:15 – 13:45
Observational Studies
Chair: Julie Funch Furberg, Novo Nordisk · Causal inference, confounding and real-world evidence
12:15 – 12:45
Target trial emulation versus real trials, concepts and challenges
Abstract
Abstract TBA — the abstract for this talk will be added once the speaker is confirmed and the programme is finalized.
12:45 – 13:15
Multicollinearity: More than inflated standard errors?
Abstract
Multicollinearity is traditionally presented as a problem of inflated standard errors and reduced precision in regression analysis. However, the challenges associated with correlated regressors may extend beyond statistical uncertainty.
In this talk, I will use examples from a PhD project investigating associations between body composition and cancer risk in a large Danish cohort. Fat mass index (FMI) and fat-free mass index (FFMI) are highly correlated, making the interpretation of their mutually adjusted associations challenging. Analyses of different cancer outcomes reveal contrasting patterns: for some outcomes, mutual adjustment attenuates associations and increases uncertainty, while for others, it produces statistically significant coefficients with unexpected signs.
Motivated by these findings, I will introduce the common-factor model, illustrating how correlated regressors can produce polarized coefficients that may not reflect the underlying structural relationships, even when the regression coefficients are statistically valid.
The talk highlights how correlated regressors can complicate the interpretation of regression coefficients, and why statistically valid estimates do not necessarily provide valid evidence about underlying biological relationships.
More broadly, the talk emphasizes the importance of considering plausible data-generating mechanisms when interpreting regression results in epidemiological research.
13:15 – 13:45
Outcome-adapted Automatic Debiased Machine Learning
Abstract
Parameters of interest in causal inference, such as treatment or policy effects, can often be expressed as linear functionals of an outcome regression function. Automatic debiased machine learning (AutoDML) is a unified framework for obtaining asymptotically normal estimators of such parameters, which requires estimation of both a regression function and a Riesz representer. Existing AutoDML neural network architectures, such as RieszNet and MADNet, use a shared intermediate covariate representation. However, it remains unclear whether this shared representation should be predictive of the Riesz representer or the outcome. We show that a shared representation of the covariates that preserves predictive power of the outcome while discarding information about the Riesz representer is asymptotically more efficient than the baseline AutoDML estimator that uses all covariates. Motivated by these results, we propose the outcome-adapted AutoDML estimator and establish its asymptotic behavior in a sample splitting framework. We provide a neural network implementation of the estimator that learns a sparse representation of the covariates that is predictive of the outcome but not predictive of the Riesz representer. We demonstrate the efficiency gains of our estimator over existing alternatives on synthetic data and achieve state-of-the-art estimation accuracy on the semi-synthetic IHDP benchmark dataset.
13:45 – 14:15
Break
Session 2 — Randomized Studies
Session 2 · 14:15 – 15:45
Randomized Studies
Chair: Birgitte Biilman Rønn, Zealand Pharma · Trial design and novel estimators
14:15 – 14:45
Prognostic versus predictive factors (effect modifiers): a superfluous distinction?
Abstract
Abstract TBA — the abstract for this talk will be added once the speaker is confirmed and the programme is finalized.
14:45 – 15:15
Things we (mis)understood about group sequential designs with multiple correlated endpoints
Abstract
In this talk we convey several obstacles to stumble upon when combining multiple testing for different endpoints with multiple testing for multiple looks from group sequential designs. Common for all the challenges were the non-trivial dependence between the type-I-error and the correlation between the tests (in multiple dimensions). We briefly introduce how group sequential designs and multiple testing work to preserve type-I-error in a randomized clinical trial setting. We then show that naively combining the two types of multiplicity control does not always lead to multiplicity control. In particular, seemingly intuitive strategies can inflate the type I error rate. Through two case studies, we illustrate situations in which intuition was not our friend and give new learned perspectives on multiple testing, correlation of tests and group sequential designs. The central lesson is simple: when you know, you know. And when you don’t, you don’t.
15:15 – 15:45
Revisiting the apparent discrepancy between the frequentist and Bayesian interpretation of an adaptive design
Abstract
Abstract TBA — the abstract for this talk will be added once the speaker is confirmed and the programme is finalized.
15:45 – 16:15
Break
Session 3 — Statisticians in an AI Era
Session 3 · 16:15 – 17:30
Statisticians in an AI Era
Chair: Anders Gorst-Rasmussen, Novo Nordisk · AI, foundation models and the role of the statistician
16:15 – 16:45
p < 0.05, Trust Me — Says the Machine
Abstract
The relevance of training in a mathematical discipline and the future roles of those that are mathematically educated have been called into question by the skills that current large language models demonstrate. These developments are not isolated to computer science and mathematics – or statistics for that matter. Concrete skills have become obsolete overnight. This will have major ramifications for education, for science and the job market and across the entire society. The question we must ask: do we want a world where no human, not even in principle, can follow and verify a decision grounded in a data analysis?
I will make a case for the intrinsic value of human learning and for humans to acquire knowledge and understanding. Not necessarily to secure a particular job, but to preserve human agency and for having a seat at the table where decisions are made. Mathematical and analytical training remains as relevant as ever as a way of sharpening our thinking about the world, even if production tasks will be increasingly taken over by AI systems.
16:45 – 17:30
Panel: The role of the Statistician in the AI era
18:30 – 22:00
Conference dinner
Day 2
Day 2 · Friday, 20th Novemeber Time-to-event · Pharmaceutical statistics
08:30 – 09:00
Coffee and arrival
Session 4 — Topics in Time-to-Event
Session 4 · 09:00 – 10:45
Topics in Time-to-Event
Chair: Henrik Ravn, Novo Nordisk · Survival analysis, competing risks and multistate models
09:00 – 09:45
Analysis of recurrent events with competing risks: a review
Abstract
In this talk, I will review two approaches to analyse recurrent events data when there are competing risks in the form of terminal events such as all-cause death. The applied framework will be that of an RCT.
The first is the classical approach based on transition intensities by which one attempts to model the full data generating process. As a consequence, one would be able to estimate all functionals, such as state occupation probabilities, sojourn times or expected number of events over time – maybe by simulations.
The second is the marginal approach – in line with the modern estimand concept. Here, an estimand (a function of the full data generating model) is targeted without the ambition of specifying that model. Typically, one would target the expected number of events over time, or the probability of having at least k events. Interpretation of such an estimand, however, is challenged by the presence of a terminal event (‘kill the patient and there will be few recurrent events’!) I will discuss two approaches to take the terminal event into account: (1) combining recurrent and terminal events into a single number to target, (2) joint analysis of estimands for recurrent events and terminal event, e.g., expected number of events and survival probabilities for the terminal event.
09:45 – 10:15
Analyzing Randomized Clinical Trials with Recurrent Events
Abstract
I will explore different approaches for estimating relevant quantities in the context of randomized clinical trials (RCT) dealing with recurrent events data. A distinctive feature of recurrent events is that the occurrences of interest are often observed and relevant only before a terminal event, such as the death of a patient. The presence of a terminal event introduces challenges in interpreting the outcomes of such events similar to what is known for competing risks.
I will consider two different outcome measures for recurrent events: the marginal mean, representing the average number of events observed before the terminal event; an intensity-based approach characterizing the rate of recurrent events among those still under risk. Simple estimators for these quantities exist. I will demonstrate how additional covariates present at the time of randomization can be used to improve the efficiency without risking modeling errors. This approach, known as RCT augmentation, is recommended when additional information is available. In cases of right censoring causing observation incompleteness, further efficiency improvement can be obtained.
10:15 – 10:45
Estimating treatment effects on duration with disease: a principal stratification framework
Abstract
Abstract TBA — the abstract for this talk will be added once the speaker is confirmed and the programme is finalized.
10:45 – 11:15
Break
Session 5 — Topics in Pharmaceutical Statistics
Session 5 · 11:15 – 12:35
Topics in Pharmaceutical Statistics
Chair: Anna Berglind, Novo Nordisk · Missing data, covariate adjustment
11:15 – 11:40
Optimizing Randomized Clinical Trials through Prognostic Score Adjustment
Abstract
In randomized clinical trials (RCTs), the accurate estimation of marginal treatment effects is crucial for determining the efficacy of interventions. Enhancing the statistical power of these analyses is a key objective for statisticians. The increasing availability of historical data from registries, prior trials, and health records presents an opportunity to improve trial efficiency. However, many methods for historical borrowing compromise strict type I error rate control. Prognostic score adjustment avoids this: a prediction model is trained on historical control data, and its predictions enter the pre-specified trial analysis as a single covariate. We present how this idea extends from linear models [Schuler et al., 2022] to two routes for marginal effect estimation. First, we add the prognostic score to the GLM plug-in estimator of Rosenblum and van der Laan. Under treatment effects additive on the link scale, this achieves local semi-parametric efficiency, and a simple conservative variance formula supports prospective power calculations. Simulations with count outcomes show smaller standard errors, nominal coverage and type I error control, even with a pure-noise score or large population shifts. Second, we incorporate the score into targeted maximum likelihood estimation (TMLE). Because TMLE is already semi-parametric efficient, the gain is not asymptotic but arises in finite samples, where a good score speeds up learning of the outcome regression and improves variance estimation.
11:40 – 12:05
Would you buy this asset? The Statistician’s role in Pharmaceutical Due Diligence
Abstract
Promising clinical trial results can make a pharmaceutical asset an attractive investment opportunity. However, a statistically significant result does not, on its own, constitute robust clinical evidence, indicate a high probability of success in future development, or ultimately make the asset an attractive investment.
Motivated by a recent pharmaceutical due diligence project, this talk will use a case study to explore the statistician’s role in evaluating a potential asset. We will examine how statistical scrutiny of the available clinical evidence can uncover risks that may not be apparent from headline results, challenge assumptions about future development, and ultimately influence the valuation of the asset.
More broadly, the case study illustrates how statisticians can contribute to pharmaceutical investment decisions by translating uncertainty in the clinical evidence into development risk and, ultimately, uncertainty in the value of the asset.
12:05 – 12:30
One-step Outcome Imputation: An Alternative to Multiple Imputation
Abstract
Missing outcomes in randomized controlled trials are often handled by multiple imputation (MI). Rubin’s rules are routinely used to estimate standard errors but can fail to provide valid standard error estimates for some commonly used procedures, such as reference-based imputation. We propose a one-step alternative by explicitly targeting the treatment effect implied by a given imputation model and constructing an efficient one-step estimator for that treatment effect via its influence function. Unlike Rubin’s rules, this approach yields asymptotically valid inference. Moreover, the proposed method circumvents the stochastic component and computational burden of MI. We illustrate the approach with examples spanning a range of imputation models, including reference-based imputation and intercurrent-event-dependent imputation.
12:35
Meeting closes - sandwiches to-go
Practical information
Program committee
- Klaus Kähler Holst (Novo Nordisk)
- Henrik Ravn (Novo Nordisk)
- Birgitte Biilmann Rønn (Zealand Pharma, DSTS)
Organizing committee
- Monica Benedita Oliveira (Novo Nordisk)
- Vibeke Frølich (Novo Nordisk)
Contact
- Email: DSBS_DSTS_EVENT@novonordisk.com
- DSTS: https://dsts.dk
- DSBS: https://dsbs.dk