Using topological data analysis and pseudo time series to infer temporal phenotypes from electronic health records

The chronic heart failure evolutions: Different fates and routes

AIMS

Individual prognostic assessment and disease evolution pathways are undefined in chronic heart failure (HF). The application of unsupervised learning methodologies could help to identify patient phenotypes and the progression in each phenotype as well as to assess adverse event risk.

METHODS AND RESULTS

From a bulk of 7948 HF patients included in the MECKI registry, we selected patients with a minimum 2-year follow-up. We implemented a topological data analysis (TDA), based on 43 variables derived from clinical, biochemical, cardiac ultrasound, and exercise evaluations, to identify several patients' clusters. Thereafter, we used the trajectory analysis to describe the evolution of HF states, which is able to identify bifurcation points, characterized by different follow-up paths, as well as specific end-stages conditions of the disease. Finally, we conducted a 5-year survival analysis (composite of cardiovascular death, left ventricular assist device, or urgent heart transplant). Findings were validated on internal (n = 527) and external (n = 777) populations. We analyzed 4876 patients (age = 63 [53-71], male gender n = 3973 (81.5%), NYHA class I-II n = 3576 (73.3%), III-IV n = 1300 (26.7%), LVEF = 33 [25.5-39.9], atrial fibrillation n = 791 (16.2%), peak VO2% pred = 54.8 [43.8-67.2]), with a minimum 2-year follow-up. Nineteen patient clusters were identified by TDA. Trajectory analysis revealed a path characterized by 3 bifurcation and 4 end-stage points. Clusters survival rate varied from 44% to 100% at 2 years and from 20% to 100% at 5 years, respectively. The event frequency at 5-year follow-up for each study cohort cluster was successfully compared with those in the validation cohorts (R = 0.94 and R = 0.84, P < 0.001, for internal and external cohort, respectively). Finally, we conducted a 5-year survival analysis (composite of cardiovascular death, left ventricular assist device, or urgent heart transplant observed in 22% of cases).

CONCLUSIONS

Each HF phenotype has a specific disease progression and prognosis. These findings allow to individualize HF patient evolutions and to tailor assessment.

Integrated Pleiotropic Gene Set Unveils Comorbidity Insights across Digestive Cancers and Other Diseases

Comorbidities are prevalent in digestive cancers, intensifying patient discomfort and complicating prognosis. Identifying potential comorbidities and investigating their genetic connections in a systemic manner prove to be instrumental in averting additional health challenges during digestive cancer management. Here, we investigated 150 diseases across 18 categories by collecting and integrating various factors related to disease comorbidity, such as disease-associated SNPs or genes from sources like MalaCards, GWAS Catalog and UK Biobank. Through this extensive analysis, we have established an integrated pleiotropic gene set comprising 548 genes in total. Particularly, there enclosed the genes encoding major histocompatibility complex or related to antigen presentation. Additionally, we have unveiled patterns in protein-protein interactions and key hub genes/proteins including TP53, KRAS, CTNNB1 and PIK3CA, which may elucidate the co-occurrence of digestive cancers with certain diseases. These findings provide valuable insights into the molecular origins of comorbidity, offering potential avenues for patient stratification and the development of targeted therapies in clinical trials.

Unraveling cradle-to-grave disease trajectories from multilayer comorbidity networks

We aim to comprehensively identify typical life-spanning trajectories and critical events that impact patients' hospital utilization and mortality. We use a unique dataset containing 44 million records of almost all inpatient stays from 2003 to 2014 in Austria to investigate disease trajectories. We develop a new, multilayer disease network approach to quantitatively analyze how cooccurrences of two or more diagnoses form and evolve over the life course of patients. Nodes represent diagnoses in age groups of ten years; each age group makes up a layer of the comorbidity multilayer network. Inter-layer links encode a significant correlation between diagnoses (p < 0.001, relative risk > 1.5), while intra-layers links encode correlations between diagnoses across different age groups. We use an unsupervised clustering algorithm for detecting typical disease trajectories as overlapping clusters in the multilayer comorbidity network. We identify critical events in a patient's career as points where initially overlapping trajectories start to diverge towards different states. We identified 1260 distinct disease trajectories (618 for females, 642 for males) that on average contain 9 (IQR 2-6) different diagnoses that cover over up to 70 years (mean 23 years). We found 70 pairs of diverging trajectories that share some diagnoses at younger ages but develop into markedly different groups of diagnoses at older ages. The disease trajectory framework can help us to identify critical events as specific combinations of risk factors that put patients at high risk for different diagnoses decades later. Our findings enable a data-driven integration of personalized life-course perspectives into clinical decision-making.

Non-stenotic fibro-calcific aortic valve as a predictor of myocardial infarction recurrence

BACKGROUND

Patients with acute myocardial infarction (AMI) are at increased risk of recurrent cardiovascular events. Non-stenotic aortic valve fibro-calcific remodeling (AVSc), reflecting systemic damage, may serve as a new marker of risk.

OBJECTIVES

To stratify subgroups of AMI patients with specific probabilities of recurrent AMI and to evaluate the importance of AVSc in this setting.

METHODS

Consecutive AMI patients (n = 2530) were admitted at Centro Cardiologico Monzino (2010-2019) and followed up for 5 years. Patients were divided into study (n = 1070) and test (n = 966) cohorts. Topological data analysis (TDA) was used to stratify patient subgroups, while Kaplan-Meier and Cox regressions analyses were used to evaluate the significance of baseline characteristics.

RESULTS

TDA identified 11 subgroups of AMI patients with specific baseline characteristics. Two subgroups showed the highest rate of reinfarction after 5 years from the indexed AMI with a combined hazard ratio (HR) of 3.8 (95%CI: 2.7-5.4) compared to the other subgroups. This was confirmed in the test cohort (HR = 3.1; 95%CI: 2.2-4.3). These two subgroups were mostly men, with hypertension and dyslipidemia, who exhibit higher prevalence of AVSc, higher levels of high-sensitive c-reactive protein and creatinine. In the year-by-year analysis, AVSc, adjusted for all confounders, showed an independent association with the increased risk of reinfarction (odds ratio of ∼2 at all time-points), in both the study and the test cohorts (all p < 0.01).

CONCLUSIONS

AVSc is a crucial variable for identifying AMI patients at high risk of recurrent AMI and its presence should be considered when assessing the management of AMI patients. The inclusion of AVSc in risk stratification models may improve the accuracy of predicting the likelihood of recurrent AMI, leading to more personalized treatment decisions.

Stratification of diabetes in the context of comorbidities, using representation learning and topological data analysis

Diabetes is a heterogenous, multimorbid disorder with a large variation in manifestations, trajectories, and outcomes. The aim of this study is to validate a novel machine learning method for the phenotyping of diabetes in the context of comorbidities. Data from 9967 multimorbid patients with a new diagnosis of diabetes were extracted from Clinical Practice Research Datalink. First, using BEHRT (a transformer-based deep learning architecture), the embeddings corresponding to diabetes were learned. Next, topological data analysis (TDA) was carried out to test how different areas in high-dimensional manifold correspond to different risk profiles. The following endpoints were considered when profiling risk trajectories: major adverse cardiovascular events (MACE), coronary artery disease (CAD), stroke (CVA), heart failure (HF), renal failure (RF), diabetic neuropathy, peripheral arterial disease, reduced visual acuity and all-cause mortality. Kaplan Meier curves were plotted for each derived phenotype. Finally, we tested the performance of an established risk prediction model (QRISK) by adding TDA-derived features. We identified four subgroups of patients with diabetes and divergent comorbidity patterns differing in their risk of future cardiovascular, renal, and other microvascular outcomes. Phenotype 1 (young with chronic inflammatory conditions) and phenotype 2 (young with CAD) included relatively younger patients with diabetes compared to phenotypes 3 (older with hypertension and renal disease) and 4 (older with previous CVA), and those subgroups had a higher frequency of pre-existing cardio-renal diseases. Within ten years of follow-up, 2592 patients (26%) experienced MACE, 2515 patients (25%) died, and 2020 patients (20%) suffered RF. QRISK3 model's AUC was augmented from 67.26% (CI 67.25-67.28%) to 67.67% (CI 67.66-67.69%) by adding specific TDA-derived phenotype and the distances to both extremities of the TDA graph improving its performance in the prediction of CV outcomes. We confirmed the importance of accounting for multimorbidity when risk stratifying heterogenous cohort of patients with new diagnosis of diabetes. Our unsupervised machine learning method improved the prediction of clinical outcomes.

The high-dimensional space of human diseases built from diagnosis records and mapped to genetic loci

Human diseases are traditionally studied as singular, independent entities, limiting researchers' capacity to view human illnesses as dependent states in a complex, homeostatic system. Here, using time-stamped clinical records of over 151 million unique Americans, we construct a disease representation as points in a continuous, high-dimensional space, where diseases with similar etiology and manifestations lie near one another. We use the UK Biobank cohort, with half a million participants, to perform a genome-wide association study of newly defined human quantitative traits reflecting individuals' health states, corresponding to patient positions in our disease space. We discover 116 genetic associations involving 108 genetic loci and then use ten disease constellations resulting from clustering analysis of diseases in the embedding space, as well as 30 common diseases, to demonstrate that these genetic associations can be used to robustly predict various morbidities.

A progression analysis of motor features in Parkinson's disease based on the mapper algorithm

Background

Parkinson's disease (PD) is a neurodegenerative disease with a broad spectrum of motor and non-motor symptoms. The great heterogeneity of clinical symptoms, biomarkers, and neuroimaging and lack of reliable progression markers present a significant challenge in predicting disease progression and prognoses.

Methods

We propose a new approach to disease progression analysis based on the mapper algorithm, a tool from topological data analysis. In this paper, we apply this method to the data from the Parkinson's Progression Markers Initiative (PPMI). We then construct a Markov chain on the mapper output graphs.

Results

The resulting progression model yields a quantitative comparison of patients' disease progression under different usage of medications. We also obtain an algorithm to predict patients' UPDRS III scores.

Conclusions

By using mapper algorithm and routinely gathered clinical assessments, we developed a new dynamic models to predict the following year's motor progression in the early stage of PD. The use of this model can predict motor evaluations at the individual level, assisting clinicians to adjust intervention strategy for each patient and identifying at-risk patients for future disease-modifying therapy clinical trials.

Analysis on Time-Series Data from Movie Using MF-DCCA Method and Recurrent Neural Network Model Under the Internet of Things

The present work aims to analyze the time-series data (TSD) from movies and support constructing the movie recommendation system. Referencing the Internet of Things (IoT) technology as the framework, a time-series data analysis system for movies is built based on the recurrent neural network (RNN) and multifractal detrended mobility cross-correlation analysis (MF-DCCA) method. First, the traditional RNN model is improved by replacing the conventional convolution operation with spatial adaptive convolution. Specifically, an additional convolution layer is used to obtain the position parameters required for adaptive convolution to improve the model performance to capture the characteristics of spatial-temporal transformation. Then, the MF-DCCA method is optimized to reduce the interference of noise signals to the analysis processing of TSD from movies. Finally, the TSD analysis system is tested for performance verification. The test results indicate that the method proposed here has outstanding stability and runs smoothly. When the prediction scheme is long short-term memory (LSTM) (L = 20), the similarity of the LSTM (L = 20) network under one frame is 0.977; the similarity of the LSTM (L = 20) network under nine frames is 0.727. This system provides a specific idea for applying the RNN model and MF-DCCA method in analyzing TSD from movies.

Topological data analysis in biomedicine: A review

Significant technological advances made in recent years have shepherded a dramatic increase in utilization of digital technologies for biomedicine- everything from the widespread use of electronic health records to improved medical imaging capabilities and the rising ubiquity of genomic sequencing contribute to a "digitization" of biomedical research and clinical care. With this shift toward computerized tools comes a dramatic increase in the amount of available data, and current tools for data analysis capable of extracting meaningful knowledge from this wealth of information have yet to catch up. This article seeks to provide an overview of emerging mathematical methods with the potential to improve the abilities of clinicians and researchers to analyze biomedical data, but may be hindered from doing so by a lack of conceptual accessibility and awareness in the life sciences research community. In particular, we focus on topological data analysis (TDA), a set of methods grounded in the mathematical field of algebraic topology that seeks to describe and harness features related to the "shape" of data. We aim to make such techniques more approachable to non-mathematicians by providing a conceptual discussion of their theoretical foundations followed by a survey of their published applications to scientific research. Finally, we discuss the limitations of these methods and suggest potential avenues for future work integrating mathematical tools into clinical care and biomedical informatics.

A Process Mining Pipeline to Characterize COVID-19 Patients' Trajectories and Identify Relevant Temporal Phenotypes From EHR Data

The impact of the COVID-19 pandemic involved the disruption of the processes of care and the need for immediately effective re-organizational procedures. In the context of digital health, it is of paramount importance to determine how a specific patients' population reflects into the healthcare dynamics of the hospital, to investigate how patients' sub-group/strata respond to the different care processes, in order to generate novel hypotheses regarding the most effective healthcare strategies. We present an analysis pipeline based on the heterogeneous collected data aimed at identifying the most frequent healthcare processes patterns, jointly analyzing them with demographic and physiological disease trajectories, and stratify the observed cohort on the basis of the mined patterns. This is a process-oriented pipeline which integrates process mining algorithms, and trajectory mining by topological data analyses and pseudo time approaches. Data was collected for 1,179 COVID-19 positive patients, hospitalized at the Italian Hospital "Istituti Clinici Salvatore Maugeri" in Lombardy, integrating different sources including text admission letters, EHR and hospital infrastructure data. We identified five temporal phenotypes, from laboratory values trajectories, which are characterized by statistically significant different death risk estimates. The process mining algorithms allowed splitting the data in sub-cohorts as function of the pandemic waves and of the temporal trajectories showing statistically significant differences in terms of events characteristics.

Dictionary learning allows model-free pseudotime estimation of transcriptomic data

Background

Pseudotime estimation from dynamic single-cell transcriptomic data enables characterisation and understanding of the underlying processes, for example developmental processes. Various pseudotime estimation methods have been proposed during the last years. Typically, these methods start with a dimension reduction step because the low-dimensional representation is usually easier to analyse. Approaches such as PCA, ICA or t-SNE belong to the most widely used methods for dimension reduction in pseudotime estimation methods. However, these methods usually make assumptions on the derived dimensions, which can result in important dataset properties being missed. In this paper, we suggest a new dictionary learning based approach, dynDLT, for dimension reduction and pseudotime estimation of dynamic transcriptomic data. Dictionary learning is a matrix factorisation approach that does not restrict the dependence of the derived dimensions. To evaluate the performance, we conduct a large simulation study and analyse 8 real-world datasets.

Results

The simulation studies reveal that firstly, dynDLT preserves the simulated patterns in low-dimension and the pseudotimes can be derived from the low-dimensional representation. Secondly, the results show that dynDLT is suitable for the detection of genes exhibiting the simulated dynamic patterns, thereby facilitating the interpretation of the compressed representation and thus the dynamic processes. For the real-world data analysis, we select datasets with samples that are taken at different time points throughout an experiment. The pseudotimes found by dynDLT have high correlations with the experimental times. We compare the results to other approaches used in pseudotime estimation, or those that are method-wise closely connected to dictionary learning: ICA, NMF, PCA, t-SNE, and UMAP. DynDLT has the best overall performance for the simulated and real-world datasets.

Conclusions

We introduce dynDLT, a method that is suitable for pseudotime estimation. Its main advantages are: (1) It presents a model-free approach, meaning that it does not restrict the dependence of the derived dimensions; (2) Genes that are relevant in the detected dynamic processes can be identified from the dictionary matrix; (3) By a restriction of the dictionary entries to positive values, the dictionary atoms are highly interpretable.

Supplementary Information

The online version contains supplementary material available at (10.1186/s12864-021-08276-9).