| 1. |
- Barros, Guilherme, 1992-
(författare)
-
Estimation of hazard ratios from observational data with applications related to stroke
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The objective of this thesis is to examine some challenges that may emerge when conducting time-to-event studies based on observational data. Time-to-event (also called survival) is a setting that involves analyzing how different factors may influence the length of time until an individual experiences the event of interest. This type of analysis is commonly applied in fields such as medical research and epidemiology. In this thesis, which focuses on stroke, we are interested in the time to a recurrent stroke or the death of a patient who survived a first stroke.Hazard ratios are one of the main parameters estimated in time-to-event studies. Hazard ratios involve comparing the risk of experiencing the event between two groups, usually a treated group and an untreated group. They can also involve other factors, such as different age groups. Hazard ratios can be estimated from the data by using the Cox regression model.Observational data, in contrast to experimental data, involves data collected without any intervention or random assignment of treatment to the individuals. Confounders, that is, variables that distort or obscure the true relationship between treatment and outcome, are always present and need to be controlled for in observational studies.National registers are an important source of observational data. A national registry is a centralized database or system that collects, stores, and maintains information about a specific population or group of individuals within a country. Sweden is known for its detailed and complete national registers. In this thesis, data from the Swedish Stroke Register (Riksstroke) is used to study factors related to stroke.In time-to-event studies involving observational data, several challenges may arise for the researcher during data analysis. Some individuals may not experience the event during the observation period and thus the information about their time until the event is incomplete. These individuals are considered as censored. Some individuals may experience another event rather than the one of interest, a competing risk. Additionally, models must be properly constructed, with researchers selecting variables and determining the suitable functional form.Four papers are included in the thesis. Paper I demonstrates how to handle competing risks in survival analysis. The study involves comparing individuals with and without standard modifiable risk factors and their risks of a recurrent stroke or death using data from the Swedish Stroke Register.The estimation of marginal hazard ratios is a common theme in the other three papers. All involve simulation studies in order to extend methods and explore best practices when estimating marginal hazard ratios.Paper II explores non-parametric methods that can be used as alternatives to more traditional parametric methods when balancing datasets in order to estimate a marginal hazard ratio. A case study was also conducted using data from the Swedish Stroke Register involving the prescription of anticoagulants at hospital discharge after a stroke.Paper III is about how censoring affects marginal hazard ratio estimation, even with perfect balancing of the dataset. We study this issue, taking into consideration varying effect sizes and censoring rates. A procedure to attenuate the problem is also studied.Paper IV concerns covariate selection in the case of high-dimensional data. High-dimensional data involves cases in which the number of covariates in the study is comparable to the number of individuals, and therefore covariate selection methods are needed. In the paper, we explore some of these methods and suggest a best-performing procedure. As Paper II, Paper IV involves a case study of anticoagulant prescription using data from the Swedish Stroke Register.
|
|
| 2. |
- Ecker, Kreske, 1987-
(författare)
-
Studying earnings trajectories as functional outcomes
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- In this thesis, we present methods for studying patterns of income accumulation over time using functional data analysis. This is made possible by the availability of large-scale longitudinal register data in Sweden. By modelling individuals’ cumulative earnings trajectories as continuous functions of time, we can explore temporal dynamics as well as divergences in these trajectories based on initial labour market conditions. A major contribution of this thesis consists of extending the potential outcome framework for causal inference to functional data analysis.In Paper I, we use functional-on-scalar linear regression and an interval-wise testing procedure to study the associations between initial labour market size and income trajectories for one Swedish birth cohort. In Paper II, we present methods to draw causal conclusions in this setting. We introduce the functional average treatment effect (FATE), as well as an outcome-regression based estimator for this parameter. In addition, we show the finite sample distribution of this estimator under certain regularity conditions and demonstrate how simultaneous confidence bands can be used for inferences about the FATE. An application study in this paper estimates the causal effect of initial labour market size on income accumulation trajectories.In Paper III, these methods are applied to study the effect of initial firm age on earnings accumulation. Paper IV presents an outcome regression based and a double robust estimator for the mean of functional outcomes when some of these outcome functions are missing at random. We derive the asymptotic distributions of these two estimators as well as their covariance structure under more general conditions.
|
|
| 3. |
- Moosavi, Niloofar, 1990-
(författare)
-
Valid causal inference in high-dimensional and complex settings
- 2022
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The objective of this thesis is to consider some challenges that arise when conducting causal inference based on observational data. High dimensionality can occur when it is necessary to adjust for many covariates, and flexible models must be used to meet convergence assumptions. The latter may require the use of a novel machine learning estimator. Estimating nonparametrically-defined causal estimands at parametric rates and obtaining good-quality confidence intervals (with near nominal coverage) are the primary goals. Another challenge is providing a sensitivity analysis that can be applied in high-dimensional scenarios as a way of assessing the robustness of the results to missing confounders. Four papers are included in the thesis. A common theme in all the papers is covariate selection or nonparametric estimation of nuisance models. To provide insight into the performance of the approaches presented, some theoretical results are provided. Additionally, simulation studies are reported. In paper I, covariate selection is discussed as a method for removing redundant variables. This approach is compared to other strategies for variable selection that ensure reasonable confidence interval coverage. Paper II integrates variable selection into a sensitivity analysis, where the sensitivity parameter is the conditional correlation of the outcome and treatment variables. The validity of the analysis where the sensitivity parameter is small relative to the sample size is shown theoretically. In simulation settings, however, the analysis performs as expected, even for larger values of sensitivity parameters, when using a correction of the estimator of the residual variance for the outcome model. Paper IV extends the applicability of the sensitivity analysis method through the use of a different residual variance estimator and applies it to a real study of the effects of smoking during pregnancy on child birth weight. A real data problem of analysing the effect of early retirement on health outcomes is studied in Paper III. Rather than using variable selection strategies, convolutional neural networks are studied to fit the nuisance models.
|
|
