» Home » Stata Conferences and Users Group meetings » 2015 Nordic and Baltic Stata Users Group meeting » Abstracts

Matteo Bottai

Karolinska Institutet

This talk describes the **grreg** command for modeling the effect of covariates on geometric rates. The occurrence of an event of interest over time is often summarized by the incidence rate, defined as the average number of events per person-time. This type of rate applies to events that may occur repeatedly over time on any given subject, such as infections. For events that can occur only once, such as death, the geometric rate may be a better summary measure.

**Additional information**

sweden15_bottai.pdf

sweden15_bottai.pdf

Hannah Bower

Karolinska Institutet

Michael J. Crowther

Karolinska Institutet and University of Leicester

Paul C. Lambert

Karolinska Institutet and University of Leicester

Flexible parametric models are frequently used as a tool to model time-to-event data. These models use restricted cubic splines to model the log cumulative baseline hazard and can be fit in Stata using the **stpm2** command. There may be certain situations where it is desirable to model on the log hazard scale rather than on the log cumulative-hazard scale. For example, modeling multiple time-dependent effects on the log cumulative-hazard scale can be problematic because of the time-dependent hazard ratio for one covariate depending on the values of other covariates; this is not the case when modeling on the log hazard scale. Modeling on the log hazard scale uses numerical integration techniques to evaluate the cumulative hazard function, which is required to maximize the likelihood. We present a new command, **strcs**, that enables implementation of flexible parametric survival models on the log hazard scale using a mixture of analytical and Gaussian quadrature integration within the estimation process, alongside many useful postestimation predictions.

**Additional information**

sweden15_bower.pdf

sweden15_bower.pdf

Viktor Oskarsson

Karolinska Institutet

Cox regression is widely used for analysis of time-to-event data. In most papers, the effect of an exposure on an outcome is reported as a single hazard ratio, with the underlying assumption that this effect is time fixed (that is, hazards are proportional on the log hazard scale). However, if the assumption of proportional hazards is violated, it is less meaningful to report a single hazard ratio; instead, a range of hazard ratios should be reported. The aims of this presentation are to (1) show how restricted cubic splines can be used in the evaluation of nonproportional hazards in Cox regression and (2) introduce a user-friendly postestimation command, **stphcoxrcs**, that greatly facilitates such numerical and graphical analysis. As an empirical example, we will study the association of fruit and vegetable consumption with risk of symptomatic gallstone disease in a prospective cohort of Swedish women.

**Additional information**

sweden15_oskarsson.pdf

sweden15_oskarsson.pdf

Yulia Marchenko

StataCorp

Stata 14 provides a suite of commands for performing Bayesian analysis. Bayesian analysis is a statistical paradigm that answers research questions about unknown parameters using probability statements. For example, what is the probability that a person accused of a crime is guilty? What is the probability that there is a positive effect of schooling on wage? What is the probability that the odds ratio is between 0.3 and 0.5? And many more. In my presentation, I will describe Stata's Bayesian suite of commands and demonstrate its use in various applications.

**Additional information**

sweden15_marchenko.pdf

sweden15_marchenko.pdf

Michele Santacatterina

Karolinska Institutet

Probability-weighted methods are commonly used in statistics to compensate for nonresponse, control for disproportional sampling fractions, balance covariate patterns, and so on. Large weights often occur in applications, causing erratic estimates and high standard errors. A common solution to this problem is to drop the observations with largest weights. This approach, however, reduces power and precision of inferences and makes it harder to interpret them. This talk describes a prototype of a new command for estimating optimal weights under specified constraints on the standard errors. The command is based on a mathematical nonlinear constrained optimization problem. Theory, implementation, and applications to real data will be presented.

**Additional information**

sweden15_santacatterina.pdf

sweden15_santacatterina.pdf

Celia García-Pareja

Karolinska Institutet

Compound expectation of an outcome variable of interest is obtained by breaking down the population into subpopulations defined by a specified set of quantiles of the outcome variable. The mean outcome is estimated in each subpopulation separately. The overall mean can be computed as a weighted sum of the components. In the presence of censoring, compound expectation can be estimated up to the last observed quantile, providing more precise inference in the subpopulations with greater number of individuals at risk. The new **cereg** command estimates the effect of a set of covariates on compound expectation of the outcome, allowing for the inclusion of numeric and categorical predictors and the evaluation of possible interactions. The use of the **cereg** command is illustrated through a real-data example.

**Additional information**

sweden15_garcia-pareja.pdf

sweden15_garcia-pareja.pdf

Giola Santoni

Karolinska Institutet

In the young or adult population, it is relatively easy to determine the health status of a person. As we get older, the concept of health becomes more a hypothetical construct that cannot be directly measured but can be derived through observable variables. Presence of more than one disease, change in cognitive functioning or in physical functioning, or appearance of some disabilities are some of the characteristic features of aging. We derive a test to evaluate the health of a person 60 years old and older using the new **irt** command (item response theory).

**Additional information**

sweden15_santoni.pdf

sweden15_santoni.pdf

Andrea Bellavia

Karolinska Institutet

Thanks to unique properties of the quantiles, statistical models for survival percentiles can be defined on the additive scale by modeling the time-to-event outcome and on the multiplicative scale by modeling its natural logarithm. The aim of this talk is to show how to use the **laplace** command to model survival percentiles on both scales. In particular, we will explore the advantages of this approach in assessing either additive or multiplicative interaction.

**Additional information**

sweden15_bellavia.pdf

sweden15_bellavia.pdf

Yulia Marchenko

StataCorp

Alan Riley

StataCorp

StataCorp will be happy to receive wishes for developments in Stata and almost as happy to
receive grumbles about the software.