Joint Modeling Strategies for Longitudinal Exposure and Time-to-Event Outcome: The Role of Childhood Nutrition and Puberty in the Development of Type 1 Diabetes

Research output: Book/ReportDoctoral thesisCollection of Articles

Abstract

Type 1 diabetes (T1D) development is a complicated process that may be initiated already in utero. The clinical diagnosis is preceded by a preclinical phase of islet autoimmunity (IA), that is, pre-T1D, in which T1D–associated autoantibodies can be detected from blood. A global rise in incidence rates during the several last decades, together with seasonal and geographical variation, suggest a critical role of environmental factors in disease development. Especially viral infections, bacteria, and diet have been considered as potential candidates. However, the overall view of the role of environmental factors is still unclear.

The aims of this thesis were 1) to design advanced statistical methods for the joint modeling of longitudinally measured childhood food consumption and the risk of pre-T1D, 2) to design statistical methods with novel approaches to combine pubertal information from three cohort studies using growth measurements, and to further model the growth-based puberty together with the development of T1D, and 3) to implement the designed methods on real data to assess the role of childhood nutrition–—consumption of meat and meat products, fish and fish products, eggs, and differently processed milk products—as well as puberty in the development of T1D.

The data originated from three Finnish prospective cohort studies: the T1D Prediction and Prevention (DIPP) Study, the Special Turku Coronary Risk Factor Intervention Project (STRIP), and the Boy cohort. All the cohorts included careful follow-up for growth from birth/infancy until adolescence. DIPP comprised children carrying increased genetic susceptibility for T1D, including regular follow-up for T1D–associated antibodies. Information on the diet was collected within DIPP Nutrition Study by using repeated 3-day food records. The STRIP and Boy cohort included a regular follow-up for pubertal development.

A standard and latent class version of a joint model for longitudinal and time-to-event data were used to study the role of childhood nutrition in the development of IA. A joint model had the ability to complete the missing data by utilizing the food consumption trajectories instead of original measurements, and smooth out a measurement error in dietary data by simultaneous fitting of the longitudinal and time-to-event submodels. Compared with the additionally fitted, naive Cox models, the hazard ratios of the joint models were further from one with wider confidence intervals, and signaled some associations.

A multi-state model was used to study the role of puberty in the development of T1D within the DIPP cohort, allowing a simultaneous estimation of hazards for both IA and further progression to T1D. Because DIPP did not include pubertal follow-up, pubertal growth markers were estimated using Super-Imposition by Translation And Rotation (SITAR) model and used as predictors in a pubertal timing assessment model constructed and validated within the STRIP and Boy cohort. A good agreement between the observed and predicted onset timings could be reached and pubertal onset timings for the DIPP children were assessed using the model. A novel function, which allowed for the study of different forms of the pubertal effect, was designed to add the effect into the multi-state model.

A determination of the components of the final models required decisions like number and places of the knots for the splines, dependence structure of the random effects in the longitudinal models, number of the groups in a joint latent class model, and the form of the pubertal effect. The decisions were mainly made based on the information criteria, together with the visual inspection. Because of flexible modeling and large data, most models became computationally intensive and convergence problems occurred. Thus, the balance between achieving the flexible solution, reasonable calculation time, and convergence of the models had to be considered when deciding the components of the models.

A higher childhood intake of cow’s milk products and meat products were (marginally) associated with an increased risk of IA, while for fish product use, the signs of a marginal protective association were detected. Egg intake was not associated with IA, and any specific milk processing type that would clearly stand out as a risk factor apart from other types could not be identified. Puberty was observed to increase a risk of progression from ICA+1-defined IA to T1D. No association between puberty and IA was detected.

The results of the present thesis strengthen the previous observations on the role of cow’s milk in the development of pre-T1D, here with a more advanced methodology within a cohort design. However, the mechanisms behind the association are still unclear. According to the present thesis, the processing of milk does not seem to play a role in association, but the finding needs confirmation. For meat, fish and eggs, a few previous prospective studies have not provided evidence on their role in T1D. For puberty, the results of this thesis raise the possibility that the previously reported peak in T1D incidence during adolescence could be linked to pubertal development. Future longitudinal studies on the role of dietary factors, among other environmental factors in different disease stages, are needed. The joint models would address the future research by allowing a simultaneous modeling of longitudinal exposures and time-to-event outcomes, also in multivariate setting. Although joint models have higher computational burden compared with naive models, they have clear advantages favoring their use over the alternatives.
Original languageEnglish
Place of PublicationTampere
ISBN (Electronic) 978-952-03-2779-8
Publication statusPublished - 2023
Publication typeG5 Doctoral dissertation (articles)

Publication series

NameTampere University Dissertations - Tampereen yliopiston väitöskirjat
Volume752
ISSN (Print)2489-9860
ISSN (Electronic)2490-0028

Fingerprint

Dive into the research topics of 'Joint Modeling Strategies for Longitudinal Exposure and Time-to-Event Outcome: The Role of Childhood Nutrition and Puberty in the Development of Type 1 Diabetes'. Together they form a unique fingerprint.

Cite this