Our empirical strategy exploits the availability of two years (January 2020 through January 2022) of daily data on the number of confirmed deaths and cases of COVID-19 in the 3,000 U.S. counties of the 48 contiguous states and the District of Columbia. In the first stage of the analysis, we use daily time-series data on COVID-19 cases and deaths to fit mixed models of deaths against lagged confirmed cases for each county. Because the resulting coefficients are county specific, they relax the homogeneity assumption that is implicit when the analysis is performed using geographically aggregated cross-section units. In the second stage of the analysis, we assume that these county estimates are a function of economic and sociodemographic factors that are taken as fixed over the course of the pandemic. Here we employ the novel one-covariate-at-a-time variable-selection algorithm proposed by Chudik et al. (2018) to guide the choice of regressors.

The current mi suite of commands for dealing with missing data does not allow a two-stage approach in fitting regression models. Therefore, I introduce a new command, twostage, that offers to fit two-stage regression models for IPD meta-analysis with missing data. twostage has been developed to accommodate systematic and sporadically missing data in IPD meta-analysis. I first briefly describe the challenges of missing data in IPD meta-analysis and then illustrate applications of the twostage command in the context of health-related studies.

The current imputation methods implemented in mi, however, are mainly suited for one study and sporadically missing data. Our aim is to introduce a new user-defined imputation method within mi impute capable of handling the main features of individual participant data (IPD) meta-analysis. Realistic simulated studies will be used to illustrate the logic and practice of imputing systematic missing data.

Recently introduced by Bottai et al. (The American Statistician. 2022. 76:4, 313–321), maximum agreement regression maximizes the concordance correlation between the prediction and the observed outcome, not the Pearson's correlation coefficient maximized by ordinary linear regression. The syntax of the command is nearly identical to that of regress, which estimates least-squares regression. The presentation shows the features of the command and its possible applications through real data examples.

To remedy the “regression to the mean” effect in randomized control trials, one should measure the continuous outcome before randomization and adjust for the baseline outcome value in the analysis. This adjustment requires the use of regression constraints. The adjustment leads to lesser standard errors. After presenting a real case, I introduce the concept of “regression to the mean.” Then I introduce the relation from “regression to the mean” to the intraclass correlation and the measurement error. Using the case, I compare the estimates from several approaches in randomized control trials. Here I demonstrate the use of constraints. Knowing the intraclass correlation in power calculations will lead to a lesser required number of observations, for example, higher power. Hence, randomized control trials should report the intraclass correlation.

These groups are known as treatment cohorts. In Stata 18, we introduced two commands that estimate treatment effects that vary over time and cohort. For repeated cross-sectional data, we have hdidregress. For panel data, we have xthdidregress. Both commands let you graph the evolution of treatment over time. They also allow you to aggregate treatment within cohort and time and visualize these effects. I will show you how both commands work and briefly discuss the theory underlying them.

A usual approach would involve fitting Cox or Poisson models to a time-split dataset. However, this leads to large datasets and can be computationally intensive when model fitting, especially if interest lies in displaying how the estimated hazard rate or survival changes along multiple time scales continuously. Flexible parametric survival models on the log-hazard scale are an alternative method when modeling data with multiple time scales. This can be achieved by using the Stata package stmt, where one of the time scales is chosen to be a primary time scale, and the other time scale(s) is(are) specified by using the offset option. Through a case study, I will demonstrate this method and provide examples of graphical representations.

In the medical field, for instance, patients are often nested within hospitals and regions, while in education, students are nested within schools. In these settings, the outcome is typically measured at the individual level, with covariates recorded at any level of the hierarchy. This hierarchical structure poses unique challenges and necessitates appropriate analytical approaches. Traditional methods, like the widely used Cox model, assume the independence of study subjects, disregarding the inherent correlations among subjects nested within the same higher-level unit (such as a hospital). Consequently, failing to account for the multilevel structure and within-cluster correlation can yield biased and inefficient results.

To address these issues, one can use mixed-effects models, which incorporate both population-level fixed effects and cluster-specific random effects at various levels of the hierarchy. Stata users can leverage several powerful commands to fit hierarchical survival models, such as mestreg and stmixed. With this presentation, I introduce and demonstrate the use of these commands, including a range of postestimation predictions. Moreover, I delve into measures that quantify the impact of the hierarchical structure, commonly referred to as contextual effects in the literature, and discuss the interpretation of model-based predictions, focusing on the difference between conditional and marginal effects.

Covariate effects are assumed to be multiplicative within both the expected hazard and the excess hazard, while the presence of disease among the studied group has an additive effect, hence the excess hazard. By modeling the expected rate, we can appropriately allow for uncertainty. The model is extended to include time-dependent effects, multiple time scales, and more. Following estimation, we quantify results through the prediction of the survival, hazard, and cumulative incidence functions, as well as transformations of these, and crucially with associated confidence intervals on all measures. The proposed method has been implemented in the Stata package stexcess (github.com/RedDoorAnalytics/stexcess).

In this presentation, I will briefly review some of the available features in Stata that have been developed over the years, with a focus towards their use in HTA, and describe some ongoing work to improve their applicability in such settings. This will include flexible survival modeling with merlin, Markov, semi-Markov and non-Markov multistate modeling with multistate, and efficient and generalizable individual patient simulation with survsim. Finally, I will introduce some new tools, such as the maic command for conducting matched-adjusted indirect comparisons, and a new prefix command for stmerlin, providing Bayesian flexible survival models.