Search bioRxiv⌕ Search

bioRxiv · 10.1101/2021.06.10.447970

Using generalized additive models to analyze biomedical non-linear longitudinal data

Abstract

1In biomedical research, the outcome of longitudinal studies has been traditionally analyzed using the repeated measures analysis of variance (rm-ANOVA) or more recently, linear mixed models (LMEMs). Although LMEMs are less restrictive than rm-ANOVA as they can work with unbalanced data and non-constant correlation between observations, both methodologies assume a linear trend in the measured response. It is common in biomedical research that the true trend response is nonlinear and in these cases the linearity assumption of rm-ANOVA and LMEMs can lead to biased estimates and unreliable inference. In contrast, GAMs relax the linearity assumption of rm-ANOVA and LMEMs and allow the data to determine the fit of the model while also permitting incomplete observations and different correlation structures. Therefore, GAMs present an excellent choice to analyze longitudinal data with non-linear trends in the context of biomedical research. This paper summarizes the limitations of rm-ANOVA and LMEMs and uses simulated data to visually show how both methods produce biased estimates when used on data with non-linear trends. We present the basic theory of GAMs and using reported trends of oxygen saturation in tumors, we simulate example longitudinal data (2 treatment groups, 10 subjects per group, 5 repeated measures for each group) to demonstrate their implementation in R. We also show that GAMs are able to produce estimates with non-linear trends even when incomplete observations exist (with 40% of the simulated observations missing). To make this work reproducible, the code and data used in this paper are available at: https://github.com/aimundo/GAMs-biomedical-research.

Source connections

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Mundo, A. I., Tipton, J. R., Muldoon, T. J.. 2021-06-11. Using generalized additive models to analyze biomedical non-linear longitudinal data. https://doi.org/10.1101/2021.06.10.447970

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

Sisyphus at the splenic flexure: sidedness in colorectal cancer

Colorectal cancer is divided into left and right to guide prognosis and therapy, yet a precise definition of left vs right sided colorectal cancer is still not fully agreed upon. The published conventions place it anywhere between the cecum and the rectosigmoid junction. The dichotomy assumes side homogeneity, i.e., that colorectal subsites within a side add limited information once the side is known. That assumption has been examined primarily through mutation prevalence, a quantity that confounds how often a driver arises with how strongly it is selected once it has. Analyzing 1,027 whole-exome and whole-genome sequenced colorectal tumors, we decomposed the prevalence of six driver groups into mutation supply and cancer effect, and asked whether a binary side label captures how each of their prevalence, mutation supply, and cancer effect vary along the colorectum. Mostly it does: the mutational input behaves as two blocks, and even the more than thirty-fold strengthening of cancer effect on BRAF V600E from sigmoid colon to cecum is adequately captured by side. But APC shows nominal evidence against side homogeneity in both prevalence and cancer effect while showing little net left-right difference, and PIK3CA H1047 rejects it in supply. The evidence for finer resolution is real but partial, and the failures do not announce themselves. We recommend additional research recording primary tumor location at the finest practicable resolution--at minimum, International Classification of Diseases for Oncology, 3rd Edition (ICD-O-3) topography. Such well-annotated datasets would enable the left-right boundary to be a reversible analytic choice rather than an irreversible annotation.

cancer biology↗

GATA2 SUMOylation Regulates GATA2 Stability to Control Uterine Serous Carcinoma Invasion

Uterine serous carcinoma (USC) is an aggressive uterine cancer subtype that rapidly invades the uterine myometrium and is responsible for 40% of uterine cancer deaths. The molecular mechanisms that underlie this aggressive phenotype are poorly understood. We reported recently that USC tumors with low expression of the zinc-finger transcription factor GATA Binding Protein 2 (GATA2) exhibit aggressive tissue invasion, while retention of GATA2 expression is an independent predictor of 100% recurrence-free and cancer-specific survival in patients with early-stage USC. Here, we explored the mechanism of GATA2 loss in USC tumors that portend invasion. We show that GATA2 in human USC cell lines has a rapid 128-minute protein half-life due to post-translational SUMOylation, which targets GATA2 for ubiquitin-dependent proteasomal degradation. Treatment with small-molecule SUMOylation inhibitors, siRNA-mediated depletion of SUMO-2/3, or mutation of GATA2 K222 and K389 increases GATA2 levels and suppresses invasion in USC cell lines and USC tumor spheroids without impacting cell proliferation or viability. Primary USC tumors from newly diagnosed patients show markedly increased SUMO-2/3 levels and upregulation of SUMOylation pathways compared to other uterine cancer subtypes or benign endometrium, and heat-shock mediated induction of USC SUMO signaling leads to a SUMOylation-dependent reduction in GATA2 levels. Our results implicate SUMOylation as a dynamic regulator of USC GATA2 levels and support a model in which stress-related increases in SUMO activity deplete GATA2 to drive USC invasion. Our findings also raise the possibility that small-molecule SUMOylation inhibitors may benefit patients with early-stage USC disease.

cancer biology↗

SLC25A39 loss disrupts mitochondrial redox homeostasis and exposes therapeutic vulnerabilities in glioblastoma

Glioblastoma is an incurable brain tumour characterised by extensive hypoxia and phenotypic plasticity, both of which drive therapy resistance. Mitochondrial glutathione, dependent on the inner membrane carrier SLC25A39, is central to buffering the oxidative stress that hypoxia generates; its regulation and significance in glioblastoma are nonetheless unknown. We find that SLC25A39 expression is hypoxia-inducible and HIF-dependent, enriched in hypoxic tumour niches and heterogeneously expressed within and between tumours. Its loss depletes mitochondrial glutathione, causing oxidative damage to nucleic acids and lipids, mitochondrial hyperfusion, impaired respiration and a compensatory shift towards glycolysis, glutamine catabolism and serine synthesis. SLC25A39-deficient cells grow poorly, yet resist temozolomide while becoming markedly sensitive to navitoclax, an inhibitor of Bcl-2 and Bcl-xL, which increases cell killing when combined with radiation and temozolomide. In this work, we identify SLC25A39 as a metabolic vulnerability and candidate stratification marker in glioblastoma.

cancer biology↗