This special issue, dedicated to Clinical Biostatistics in the 2020s, includes peer-reviewed manuscripts that reflect novel results of biostatistical research presented at the 41st Annual Conference of the International Society for Clinical Biostatistics (ISCB), hosted by the 650-year-old Jagellonian University (UJ) in Krakow, Poland, on August 23–27, 2020. This was the first virtual ISCB conference. The Local Organizing Committee (LOC) was chaired by Krystyna Szafraniec from UJ, in Krakow, Poland. The Scientific Program Committee (SPC) was chaired by Michal Abrahamowicz from McGill University in Montreal, Canada, and included 20 experts in different areas of statistics, from all six continents. The decision to change the conference format to virtual, forced by the Covid-19 pandemic and taken about 3 months before the opening, created big challenges specific to the (then quite recent) pandemic context and required creative solutions for both the LOC and SPC. Whereas the virtual program structure was similar to traditional in-person ISCB conferences, the entire time schedule of all sessions had to be reorganized to respect the 16 different time zones of participants, while ensuring the congruency of topics presented by individual speakers. In spite of these unusual challenges, the 2020 conference attracted 750 online participants from 51 countries on all six continents, making it one of the largest meetings in the ISCB history. The scientific program attempted to cover state-of-the-art methodological developments in a vast spectrum of biostatistical research. Plenary sessions included the ISCB President's Invited address by Miguel A. Hernan (Harvard) on Using causal inference to learn from real-world evidence, and the Keynote Session, organized as a discussion on Statistical modelling versus machine learning between Frank Harrell Jr. (Vanderbilt) and Trevor Hastie (Stanford). The 13 invited sessions involved 40 presenters (18 women and 22 men) from 18 countries on five continents, who discussed recent advances in both traditional areas of biostatistics (causal inference, clinical trials, innovative designs, longitudinal data, measurement errors and other data deficiencies, survival analysis), and emerging fields (big data and functional analysis, omics and personalized medicine, Bayesian modeling of infectious diseases, STRATOS initiative, and statistical challenges in Covid-19 research). From more than 450 submitted abstracts, 210 were selected for oral contributed talks, and 211 for posters (including 88 presented in 5-min overview talks). Five preconference courses and three minisymposia, attended by, respectively, a total of 220 and 270 participants, completed the scientific program. The 26 manuscripts submitted for this special issue were evaluated by six guest associate editors, who all were also SPC members: Michal Abrahamowicz (McGill University, Montreal, Canada), Yingying Fan (University of Southern California, Los Angeles, USA), Krista Fischer (University of Tartu, Estonia), Ruth Keogh (London School of Hygiene & Tropical Medicine, UK), Willi Sauerbrei (University of Freiburg, Germany), and Toshiro Tango (University of Tokyo, Japan). Through a highly competitive review process, seven manuscripts were accepted for publication. Gregorio et al. address challenges at the intersection of causal inference and survival analysis. Specifically, they extend marginal structural models (MSMs) to complex time-to-event analyses involving recurrent events which are correlated with the hazard of mortality endpoint. They address analytical challenges related to both joint modeling and flexible modeling necessary to account for potential time-varying effects of dynamic treatments. They also discuss how to define relevant causal estimands in this complex setting. Leger et al. focus on a different complexity often encountered in real-world applications of causal inference, that is near violation of the underlying positivity assumption. This problem occurs when the estimated probability of exposure is very close to either 0 or 1 for some observed vectors of covariates included in the model for estimating inverse probability weights. They compare the ability of alternative methods to correct for the violation of this crucial assumption. The results of their simulations demonstrate both the importance of accounting for this frequent problem, and the encouraging performance of g-estimation and targeted maximum likelihood methods in this context. Wijesuriya et al. address another frequent limitation of real-world studies, that is the need for imputing missing data, and extend the existing multiple imputation (MI) methodology to complex multilevel modeling of longitudinal studies. In particular, they adapt the substantive model compatible MI approach to repeated within-person measures of doubly clustered data, resulting in three levels of clustering. This novel method is then shown to efficiently correct for missing data in complex analyses involving nonlinear or time-dependent exposure effects, or time-varying covariates and their interactions. Hackenberg et al. are also interested in a specific data imperfection that often occurs in longitudinal studies if the within-patient repeated measures are sparse. They propose a novel combination of ordinary differential equations for dynamic modeling with deep learning algorithms that exploit between-patient similarities. This ingenious approach allows them to accurately estimate latent longitudinal trajectories for individual patients even if their observed data are limited to just two time points. Faucheux et al. also address the issue of sparse longitudinal measurements, but in the different context of estimating the relationships between patterns of changes in sparsely measured markers of immunity and relapse-free survival in breast cancer. They explore an innovative semisupervised learning approach, which aims to achieve multiobjective optimization that combines unsupervised and supervised learning. The former involves MI, while the latter incorporates additional information on survival, yielded by cross-validation. Abrahamowicz et al. tackle yet another practical limitation of data often encountered in real-world applications of survival analysis. They focus on interval-censored events, that is endpoints whose occurrence can be established through a clinical assessment, at discrete, often sparse, times of clinic visits. They first demonstrate that interval censoring induces serious bias to the null in the estimated effects of time-varying treatments or exposures, such as cumulative duration or dose of a drug treatment. Then, they propose a novel adaptation of the simulation-extrapolation (SIMEX) methodology to this setting and show in simulations that it largely reduces the bias. Ursino et al. focus also on the accurate assessment of the effects of a cumulative dose of a drug, but in the different setting of phase I dose-finding trials of new cancer treatments. They propose a new Bayesian method for estimating the impact of cumulative dose on the risk of toxicity adverse events. In simulations, they demonstrate the efficiency of their novel approach and its usefulness for prediction. Almost all articles in this issue propose novel statistical methods to address different complex analytical challenges highly relevant for modern clinical biostatistics. All seven papers include also both carefully designed simulations to assess or compare the methods of interest and real-world applications that illustrate their potential practical advantages. Importantly, all authors adhere to the Biometrical Journal’s high standards for reproducible research, and all provide the independently tested) programs that allows implementing their respective methods and reproducing the results presented in the articles. Finally, the fact that the first authors of the seven papers represent three different continents reflects the global outreach of the conference. Following recent trends in biostatistical research, many of the proposed novel methods combine different approaches developed earlier in separate areas of statistical research, for example, survival analysis versus causal inference or measurement errors. Whereas each of the seven articles addresses a different analytical challenge and, thus, proposes a different methodology, many themes of high relevance to modern biostatistics are common to several papers, as illustrated in Table 1. For example, many authors address specific limitations or imperfections of data typically analyzed in real-world clinical or epidemiological studies, such as missing data or sparse longitudinal measurements; or address additional complexities induced by innovative designs, such as multilevel clustering or dose-finding trials. As expected, different approaches to machine learning, including semisupervised and deep learning, and other computing-intense methods, as well as the analysis of longitudinal studies, with time-varying treatments or exposures, are increasingly represented among the articles published in this special issue. This selection of themes fits well with the ISCB 2020′s scientific program, where dedicated invited sessions focused on each of the aforementioned challenges.