Spencer Farrell
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Spencer Farrell has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Andrew D. Rutenberg
Department of Physics and Atmospheric Science, Dalhousie University, Halifax, Nova Scotia, Canada
10 shared publications Arnold Mitnitski
Affiliations Division of Geriatric Medicine, Dalhousie University, Halifax, Nova Scotia, Canada, Department of Medicine, Dalhousie University, Halifax, Nova Scotia, Canada
6 shared publicationsKenneth Rockwood
Division of Geriatric Medicine Dalhousie University B3H 2E1 Halifax Nova Scotia Canada
5 shared publications- Kenneth Rockwood
Division of Geriatric Medicine, Dalhousie University , Halifax, Nova Scotia
5 shared publications - Arnold Mitnitski
Dalhousie University
3 shared publications - Olga Theou
Geriatric Medicine Research, Nova Scotia Health & Dalhousie University, Halifax, Nova Scotia, Canada; School of Physiotherapy, Dalhousie University, Halifax, Nova Scotia, Canada
3 shared publications
Latest Publications
- Open Access
- 10 January 2022
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 18 (1) , e1009746
Interpretable machine learning for high-dimensional trajectories of aging health
We have built a computational model for individual aging trajectories of health and survival, which contains physical, functional, and biological variables, and is conditioned on demographic, lifestyle, and medical background information. We combine techniques of modern machine learning with an interpretable interaction network, where health variables are coupled by explicit pair-wise interactions within a stochastic dynamical system. Our dynamic joint interpretable network (DJIN) model is scalable to large longitudinal data sets, is predictive of individual high-dimensional health trajectories and survival from baseline health states, and infers an interpretable network of directed interactions between the health variables. The network identifies plausible physiological connections between health variables as well as clusters of strongly connected health variables. We use English Longitudinal Study of Aging (ELSA) data to train our model and show that it performs better than multiple dedicated linear models for health outcomes and survival. We compare our model with flexible lower-dimensional latent-space models to explore the dimensionality required to accurately model aging health outcomes. Our DJIN model can be used to generate synthetic individuals that age realistically, to impute missing data, and to simulate future aging outcomes given arbitrary initial health states.
- 10 January 2022
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 18 (1) , e1009746
Open AccessInterpretable machine learning for high-dimensional trajectories of aging health
We have built a computational model for individual aging trajectories of health and survival, which contains physical, functional, and biological variables, and is conditioned on demographic, lifestyle, and medical background information. We combine techniques of modern machine learning with an interpretable interaction network, where health variables are coupled by explicit pair-wise interactions within a stochastic dynamical system. Our dynamic joint interpretable network (DJIN) model is scalable to large longitudinal data sets, is predictive of individual high-dimensional health trajectories and survival from baseline health states, and infers an interpretable network of directed interactions between the health variables. The network identifies plausible physiological connections between health variables as well as clusters of strongly connected health variables. We use English Longitudinal Study of Aging (ELSA) data to train our model and show that it performs better than multiple dedicated linear models for health outcomes and survival. We compare our model with flexible lower-dimensional latent-space models to explore the dimensionality required to accurately model aging health outcomes. Our DJIN model can be used to generate synthetic individuals that age realistically, to impute missing data, and to simulate future aging outcomes given arbitrary initial health states.
- Open Access
Interpretable machine learning for high-dimensional trajectories of aging health
We have built a computational model for individual aging trajectories of health and survival, which contains physical, functional, and biological variables, and is conditioned on demographic, lifestyle, and medical background information. We combine techniques of modern machine learning with an interpretable interaction network, where health variables are coupled by explicit pair-wise interactions within a stochastic dynamical system. Our dynamic joint interpretable network (DJIN) model is scalable to large longitudinal data sets, is predictive of individual high-dimensional health trajectories and survival from baseline health states, and infers an interpretable network of directed interactions between the health variables. The network identifies plausible physiological connections between health variables as well as clusters of strongly connected health variables. We use English Longitudinal Study of Aging (ELSA) data to train our model and show that it performs better than multiple dedicated linear models for health outcomes and survival. We compare our model with flexible lower-dimensional latent-space models to explore the dimensionality required to accurately model aging health outcomes. Our DJIN model can be used to generate synthetic individuals that age realistically, to impute missing data, and to simulate future aging outcomes given arbitrary initial health states.
Open AccessInterpretable machine learning for high-dimensional trajectories of aging health
We have built a computational model for individual aging trajectories of health and survival, which contains physical, functional, and biological variables, and is conditioned on demographic, lifestyle, and medical background information. We combine techniques of modern machine learning with an interpretable interaction network, where health variables are coupled by explicit pair-wise interactions within a stochastic dynamical system. Our dynamic joint interpretable network (DJIN) model is scalable to large longitudinal data sets, is predictive of individual high-dimensional health trajectories and survival from baseline health states, and infers an interpretable network of directed interactions between the health variables. The network identifies plausible physiological connections between health variables as well as clusters of strongly connected health variables. We use English Longitudinal Study of Aging (ELSA) data to train our model and show that it performs better than multiple dedicated linear models for health outcomes and survival. We compare our model with flexible lower-dimensional latent-space models to explore the dimensionality required to accurately model aging health outcomes. Our DJIN model can be used to generate synthetic individuals that age realistically, to impute missing data, and to simulate future aging outcomes given arbitrary initial health states.
- 1 January 2021
- Published by Elsevier in Mechanisms of Ageing and Development
- Vol. 193, 111403
The potential for complex computational models of aging
- 1 January 2021
- Published by Elsevier in Mechanisms of Ageing and Development
- Vol. 193, 111403
The potential for complex computational models of aging
- Open Access
- 16 November 2020
- Published by Springer Nature in Scientific Reports
- Vol. 10 (1) , 1-11
Generating synthetic aging trajectories with a weighted network model using cross-sectional data
We develop a computational model of human aging that generates individual health trajectories with a set of observed health attributes. Our model consists of a network of interacting health attributes that stochastically damage with age to form health deficits, leading to eventual mortality. We train and test the model for two different cross-sectional observational aging studies that include simple binarized clinical indicators of health. In both studies, we find that cohorts of simulated individuals generated from the model resemble the observed cross-sectional data in both health characteristics and mortality. We can generate large numbers of synthetic individual aging trajectories with our weighted network model. Predicted average health trajectories and survival probabilities agree well with the observed data.
- 16 November 2020
- Published by Springer Nature in Scientific Reports
- Vol. 10 (1) , 1-11
Open AccessGenerating synthetic aging trajectories with a weighted network model using cross-sectional data
We develop a computational model of human aging that generates individual health trajectories with a set of observed health attributes. Our model consists of a network of interacting health attributes that stochastically damage with age to form health deficits, leading to eventual mortality. We train and test the model for two different cross-sectional observational aging studies that include simple binarized clinical indicators of health. In both studies, we find that cohorts of simulated individuals generated from the model resemble the observed cross-sectional data in both health characteristics and mortality. We can generate large numbers of synthetic individual aging trajectories with our weighted network model. Predicted average health trajectories and survival probabilities agree well with the observed data.
- Open Access
The potential for complex computational models of aging
The gradual accumulation of damage and dysregulation during the aging of living organisms can be quantified. Even so, the aging process is complex and has multiple interacting physiological scales -- from the molecular to cellular to whole tissues. In the face of this complexity, we can significantly advance our understanding of aging with the use of computational models that simulate realistic individual trajectories of health as well as mortality. To do so, they must be systems-level models that incorporate interactions between measurable aspects of age-associated changes. To incorporate individual variability in the aging process, models must be stochastic. To be useful they should also be predictive, and so must be fit or parameterized by data from large populations of aging individuals. In this perspective, we outline where we have been, where we are, and where we hope to go with such computational models of aging. Our focus is on data-driven systems-level models, and on their great potential in aging research.
Open AccessThe potential for complex computational models of aging
The gradual accumulation of damage and dysregulation during the aging of living organisms can be quantified. Even so, the aging process is complex and has multiple interacting physiological scales -- from the molecular to cellular to whole tissues. In the face of this complexity, we can significantly advance our understanding of aging with the use of computational models that simulate realistic individual trajectories of health as well as mortality. To do so, they must be systems-level models that incorporate interactions between measurable aspects of age-associated changes. To incorporate individual variability in the aging process, models must be stochastic. To be useful they should also be predictive, and so must be fit or parameterized by data from large populations of aging individuals. In this perspective, we outline where we have been, where we are, and where we hope to go with such computational models of aging. Our focus is on data-driven systems-level models, and on their great potential in aging research.
- Open Access
Generating synthetic aging trajectories with a weighted network model using cross-sectional data
We develop a computational model of human aging that generates individual health trajectories with a set of observed health attributes. Our model consists of a network of interacting health attributes that stochastically damage with age to form health deficits, leading to eventual mortality. We train and test the model for two different cross-sectional observational aging studies that include simple binarized clinical indicators of health. In both studies, we find that cohorts of simulated individuals generated from the model resemble the observed cross-sectional data in both health characteristics and mortality. We can generate large numbers of synthetic individual aging trajectories with our weighted network model. Predicted average health trajectories and survival probabilities agree well with the observed data.
Open AccessGenerating synthetic aging trajectories with a weighted network model using cross-sectional data
We develop a computational model of human aging that generates individual health trajectories with a set of observed health attributes. Our model consists of a network of interacting health attributes that stochastically damage with age to form health deficits, leading to eventual mortality. We train and test the model for two different cross-sectional observational aging studies that include simple binarized clinical indicators of health. In both studies, we find that cohorts of simulated individuals generated from the model resemble the observed cross-sectional data in both health characteristics and mortality. We can generate large numbers of synthetic individual aging trajectories with our weighted network model. Predicted average health trajectories and survival probabilities agree well with the observed data.
- Open Access
Informative Frailty Indices from Binarized Biomarkers
Frailty indices (FI) based on continuous valued health data, such as obtained from blood and urine tests, have been shown to be predictive of adverse health outcomes. However, creating FI from such biomarker data requires a binarization treatment that is difficult to standardize across studies. In this work, we explore a “quantile” methodology for the generic treatment of biomarker data that allows us to construct an FI without preexisting medical knowledge (i.e. risk thresholds) of the included biomarkers. We show that our quantile approach performs as well as, or even slightly better than, established methods for the National Health and Nutrition Examination Survey (NHANES) and the Canadian Study of Health and Aging (CSHA) data sets. Furthermore, we show that our approach is robust to cohort effects within studies as compared to other data-based methods. The success of our binarization approaches provides insight into the robustness of the FI as a health measure, the upper limits of the FI observed in various data sets, and highlights general difficulties in obtaining absolute scales for comparing FI between studies.
Open AccessInformative Frailty Indices from Binarized Biomarkers
Frailty indices (FI) based on continuous valued health data, such as obtained from blood and urine tests, have been shown to be predictive of adverse health outcomes. However, creating FI from such biomarker data requires a binarization treatment that is difficult to standardize across studies. In this work, we explore a “quantile” methodology for the generic treatment of biomarker data that allows us to construct an FI without preexisting medical knowledge (i.e. risk thresholds) of the included biomarkers. We show that our quantile approach performs as well as, or even slightly better than, established methods for the National Health and Nutrition Examination Survey (NHANES) and the Canadian Study of Health and Aging (CSHA) data sets. Furthermore, we show that our approach is robust to cohort effects within studies as compared to other data-based methods. The success of our binarization approaches provides insight into the robustness of the FI as a health measure, the upper limits of the FI observed in various data sets, and highlights general difficulties in obtaining absolute scales for comparing FI between studies.
- 10 September 2018
- Published by American Physical Society (APS) in Physical Review E
- Vol. 98 (3) , 032302
Probing the network structure of health deficits in human aging
We confront a network model of human aging and mortality in which nodes represent health attributes that interact within a scale-free network topology, with observational data that use both clinical and laboratory (preclinical) health deficits as network nodes. We find that individual health attributes exhibit a wide range of mutual information with mortality and that, with a reconstruction of their relative connectivity, higher-ranked nodes are more informative. Surprisingly, we find a broad and overlapping range of mutual information of laboratory measures as compared with clinical measures. We confirm similar behavior between most-connected and least-connected model nodes, controlled by the nearest-neighbor connectivity. Furthermore, in both model and observational data, we find that the least-connected (laboratory) nodes are damaged earlier than the most-connected (clinical) deficits. A mean-field theory of our network model captures and explains this phenomenon, which results from the connectivity of nodes and of their connected neighbors. We find that other network topologies, including random, small-world, and assortative scale-free networks, exhibit qualitatively different behavior. Our disassortative scale-free network model behaves consistently with our expanded phenomenology observed in human aging and so is a useful tool to explore mechanisms of and to develop predictive measures for human aging and mortality.
- 10 September 2018
- Published by American Physical Society (APS) in Physical Review E
- Vol. 98 (3) , 032302
Probing the network structure of health deficits in human aging
We confront a network model of human aging and mortality in which nodes represent health attributes that interact within a scale-free network topology, with observational data that use both clinical and laboratory (preclinical) health deficits as network nodes. We find that individual health attributes exhibit a wide range of mutual information with mortality and that, with a reconstruction of their relative connectivity, higher-ranked nodes are more informative. Surprisingly, we find a broad and overlapping range of mutual information of laboratory measures as compared with clinical measures. We confirm similar behavior between most-connected and least-connected model nodes, controlled by the nearest-neighbor connectivity. Furthermore, in both model and observational data, we find that the least-connected (laboratory) nodes are damaged earlier than the most-connected (clinical) deficits. A mean-field theory of our network model captures and explains this phenomenon, which results from the connectivity of nodes and of their connected neighbors. We find that other network topologies, including random, small-world, and assortative scale-free networks, exhibit qualitatively different behavior. Our disassortative scale-free network model behaves consistently with our expanded phenomenology observed in human aging and so is a useful tool to explore mechanisms of and to develop predictive measures for human aging and mortality.
- Open Access
Probing the network structure of health deficits in human aging
We confront a network model of human aging and mortality in which nodes represent health attributes that interact within a scale-free network topology, with observational data that uses both clinical and laboratory (pre-clinical) health deficits as network nodes. We find that individual health attributes exhibit a wide range of mutual information with mortality and that, with a re- construction of their relative connectivity, higher-ranked nodes are more informative. Surprisingly, we find a broad and overlapping range of mutual information of laboratory measures as compared with clinical measures. We confirm similar behavior between most-connected and least-connected model nodes, controlled by the nearest-neighbor connectivity. Furthermore, in both model and observational data, we find that the least-connected (laboratory) nodes damage earlier than the most-connected (clinical) deficits. A mean-field theory of our network model captures and explains this phenomenon, which results from the connectivity of nodes and of their connected neighbors. We find that other network topologies, including random, small-world, and assortative scale-free net- works, exhibit qualitatively different behavior. Our disassortative scale-free network model behaves consistently with our expanded phenomenology observed in human aging, and so is a useful tool to explore mechanisms of and to develop new predictive measures for human aging and mortality.
Open AccessProbing the network structure of health deficits in human aging
We confront a network model of human aging and mortality in which nodes represent health attributes that interact within a scale-free network topology, with observational data that uses both clinical and laboratory (pre-clinical) health deficits as network nodes. We find that individual health attributes exhibit a wide range of mutual information with mortality and that, with a re- construction of their relative connectivity, higher-ranked nodes are more informative. Surprisingly, we find a broad and overlapping range of mutual information of laboratory measures as compared with clinical measures. We confirm similar behavior between most-connected and least-connected model nodes, controlled by the nearest-neighbor connectivity. Furthermore, in both model and observational data, we find that the least-connected (laboratory) nodes damage earlier than the most-connected (clinical) deficits. A mean-field theory of our network model captures and explains this phenomenon, which results from the connectivity of nodes and of their connected neighbors. We find that other network topologies, including random, small-world, and assortative scale-free net- works, exhibit qualitatively different behavior. Our disassortative scale-free network model behaves consistently with our expanded phenomenology observed in human aging, and so is a useful tool to explore mechanisms of and to develop new predictive measures for human aging and mortality.
- Open Access
Probing the network structure of health deficits in human aging
Human aging leads to the stochastic accumulation of damage. We model an aging population using a stochastic network model. Individuals are modeled as a network of interacting nodes, representing health attributes. Nodes in the network stochastically damage and repair, with rates dependent on the state of their neighbors. Damaged nodes represent health deficits. The Frailty Index (FI) assesses age-related damage as the proportion of health deficits an individual has accumulated, from a selection of attributes. Here, we use computational, information-theoretic, and mean-field approaches to show that the degree distribution and degree correlations of the underlying network are important to the model’s ability to recover the behavior of observational health data. We use different measures of damage in the network to probe the structure of the network. We find that the behavior of different classes of observational health deficits (laboratory or clinical) is similar to the behavior of nodes of low or high degree in the model, respectively. This explains how damage can propagate within the network, leading towards individual mortality.
Open AccessProbing the network structure of health deficits in human aging
Human aging leads to the stochastic accumulation of damage. We model an aging population using a stochastic network model. Individuals are modeled as a network of interacting nodes, representing health attributes. Nodes in the network stochastically damage and repair, with rates dependent on the state of their neighbors. Damaged nodes represent health deficits. The Frailty Index (FI) assesses age-related damage as the proportion of health deficits an individual has accumulated, from a selection of attributes. Here, we use computational, information-theoretic, and mean-field approaches to show that the degree distribution and degree correlations of the underlying network are important to the model’s ability to recover the behavior of observational health data. We use different measures of damage in the network to probe the structure of the network. We find that the behavior of different classes of observational health deficits (laboratory or clinical) is similar to the behavior of nodes of low or high degree in the model, respectively. This explains how damage can propagate within the network, leading towards individual mortality.