Efficiently comprehending diagnosis and treatment plans remains a significant challenge for both medical professionals and patients, particularly when dealing with rare or newly emerging diseases and specific combinations of comorbidities. We present MedAssist, a large language model (LLM)-empowered medical assistant designed to support the scrutinization and comprehension of electronic health records (EHRs). MedAssist leverages two key components: medical knowledge retrieval, which retrieves the latest and most comprehensive medical knowledge snippets from the web, and data retrieval, which extracts diagnosis and treatment plans for similar patients from existing EHR databases. By integrating these capabilities into user-friendly interfaces, MedAssist bridges critical gaps in medical knowledge accessibility and understanding, and advances patient care in realistic clinical scenarios.
The rapid growth of scientific publications and evolving experimental paradigms create significant challenges in staying up-to-date with current advances. Assertions are often unstructured and have limited provenance, which hinders reproducibility. Ontologies and knowledge graphs (KGs) offer structured solutions by capturing assertions, evidence, and provenance to support reproducibility. This paper reviews 23 ontologies - 13 focused on assertions and evidence and 10 on provenance - providing an overview of the current landscape while highlighting key challenges and opportunities for improvement.
Graph Neural Networks (GNNs) have achieved great success in learning with graph-structured data. Privacy concerns have also been raised for the trained models which could expose the sensitive information of graphs including both node features and the structure information. In this paper, we aim to achieve node-level differential privacy (DP) for training GNNs so that a node and its edges are protected. Node DP is inherently difficult for GNNs because all direct and multi-hop neighbors participate in the calculation of gradients for each node via layer-wise message passing and there is no bound on how many direct and multi-hop neighbors a node can have, so existing DP methods will result in high privacy cost or poor utility due to high node sensitivity. We propose a D ecoupled GNN with Differentially P rivate A pproximate Personalized PageR ank (DPAR) for training GNNs with an enhanced privacy-utility tradeoff. The key idea is to decouple the feature projection and message passing via a DP PageRank algorithm which learns the structure information and uses the top-K neighbors determined by the PageRank for feature aggregation. By capturing the most important neighbors for each node and avoiding the layer-wise message passing, it bounds the node sensitivity and achieves improved privacy-utility tradeoff compared to layer-wise perturbation based methods. We theoretically analyze the node DP guarantee for the two processes combined together and empirically demonstrate better utilities of DPAR with the same level of node DP compared with state-of-the-art methods.
Patient risk prediction models are crucial as they enable healthcare providers to proactively identify and address potential health risks. Large pre-trained foundation models offer remarkable performance in risk prediction tasks by analyzing multimodal patient data. However, a notable limitation of pre-trained foundation models lies in their deterministic predictions (i.e., lacking the ability to acknowledge uncertainty). We propose Gaussian Process-based foundation models to enable the generation of accurate predictions with instance-level uncertainty quantification, thus allowing healthcare professionals to make more informed and cautious decisions. Our proposed approach is principled and architecture-agnostic. Experimental results show that our proposed approach achieves competitive performance on classical classification metrics. Moreover, we observe that the accuracy of certain predictions is much higher than that of the uncertain ones, which validates the uncertainty awareness of our proposed method. Therefore, healthcare providers can trust low-uncertainty predictions and conduct more comprehensive investigations on high-uncertainty predictions, ultimately enhancing patient outcomes with less expert intervention.
Model card reports provide a transparent description of machine learning models which includes information about their evaluation, limitations, intended use, etc. Federal health agencies have expressed an interest in model cards report for research studies using machine-learning based AI. Previously, we have developed an ontology model for model card reports to structure and formalize these reports. In this paper, we demonstrate a Java-based library (OWL API, FaCT++) that leverages our ontology to publish computable model card reports. We discuss future directions and other use cases that highlight applicability and feasibility of ontology-driven systems to support FAIR challenges.
Data from Web artifacts and from the Web is often sensitive and cannot be directly shared for data analysis. Therefore, synthetic data generated from the real data is increasingly used as a privacy-preserving substitute. In many cases, real data from the web has missing values where the missingness itself possesses important informational content, which domain experts leverage to improve their analysis. However, this information content is lost if either imputation or deletion is used before synthetic data generation. In this paper, we propose several methods to generate synthetic data that preserve both the observable and the missing data distributions. An extensive empirical evaluation over a range of carefully fabricated and real world datasets demonstrates the effectiveness of our approach.
Federated learning (FL) aims to train powerful and generalized global models without putting distributed data together, which has been shown effective in various domains of machine learning. The non-IIDness of data across local clients has been a major challenge for FL. In graphs, one specifically important perspective of non-IIDness is manifested in the link-type heterogeneity underlying homogeneous graphs– the seemingly uniform links captured in most real-world networks can carry different levels of homophily or semantics of relations, while the exact sets and distributions of such latent link-types can further differ across local clients. Through our preliminary data analysis, we are motivated to design a new graph FL framework that can simultaneously discover latent link-types and model message-passing w.r.t. the discovered link-types through the collaboration of distributed local clients. Specifically, we propose a framework FedLit that can dynamically detect the latent link-types during FL via an EM-based clustering algorithm and differentiate the message-passing through different types of links via multiple convolution channels. For experiments, we synthesize multiple realistic datasets of graphs with latent heterogeneous link-types from real-world data, and partition them with different levels of link-type heterogeneity. Comprehensive experimental results and in-depth analysis have demonstrated both superior performance and rational behaviors of our proposed techniques.
Since the rise of the COVID-19 pandemic, peer-reviewed biomedical repositories have experienced a surge in chemical and disease related queries. These queries have a wide variety of naming conventions and nomenclatures from trademark and generic, to chemical composition mentions. Normalizing or disambiguating these mentions within texts provides researchers and data-curators with more relevant articles returned by their search query. Named entity normalization aims to automate this disambiguation process by linking entity mentions onto their appropriate candidate concepts within a biomedical knowledge base or ontology. We explore several term embedding aggregation techniques in addition to how the term's context affects evaluation performance. We also evaluate our embedding approaches for normalizing term instances containing one or many relations within unstructured texts.
The Unified Medical Language System (UMLS) Metathesaurus construction process mainly relies on lexical algorithms and manual expert curation for integrating over 200 biomedical vocabularies. A lexical-based learning model (LexLM) was developed to predict synonymy among Metathesaurus terms and largely outperforms a rule-based approach (RBA) that approximates the current construction process. However, the LexLM has the potential for being improved further because it only uses lexical information from the source vocabularies, while the RBA also takes advantage of contextual information. We investigate the role of multiple types of contextual information available to the UMLS editors, namely source synonymy (SS), source semantic group (SG), and source hierarchical relations (HR), for the UMLS vocabulary alignment (UVA) problem. In this paper, we develop multiple variants of context-enriched learning models (ConLMs) by adding to the LexLM the types of contextual information listed above. We represent these context types in context-enriched knowledge graphs (ConKGs) with four variants ConSS, ConSG, ConHR, and ConAll. We train these ConKG embeddings using seven KG embedding techniques. We create the ConLMs by concatenating the ConKG embedding vectors with the word embedding vectors from the LexLM. We evaluate the performance of the ConLMs using the UVA generalization test datasets with hundreds of millions of pairs. Our extensive experiments show a significant performance improvement from the ConLMs over the LexLM, namely +5.0% in precision (93.75%), +0.69% in recall (93.23%), +2.88% in F1 (93.49%) for the best ConLM. Our experiments also show that the ConAll variant including the three context types takes more time, but does not always perform better than other variants with a single context type. Finally, our experiments show that the pairs of terms with high lexical similarity benefit most from adding contextual information, namely +6.56% in precision (94.97%), +2.13% in recall (93.23%), +4.35% in F1 (94.09%) for the best ConLM. The pairs with lower degrees of lexical similarity also show performance improvement with +0.85% in F1 (96%) for low similarity and +1.31% in F1 (96.34%) for no similarity. These results demonstrate the importance of using contextual information in the UVA problem.
Modern healthcare systems knitted by a web of entities (e.g., hospitals, clinics, pharmacy companies) are collecting a huge volume of healthcare data from a large number of individuals with various medical procedures, medications, diagnosis, and lab tests. To extract meaningful medical concepts (i.e., phenotypes) from such higher-arity relational healthcare data, tensor factorization has been proven to be an effective approach and received increasing research attention, due to their intrinsic capability to represent the high-dimensional data. Recently, federated learning offers a privacy-preserving paradigm for collaborative learning among different entities, which seemingly provides an ideal potential to further enhance the tensor factorization-based collaborative phenotyping to handle sensitive personal health data. However, existing attempts to federated tensor factorization come with various limitations, including restrictions to the classic tensor factorization, high communication cost and reduced accuracy. We propose a communication efficient federated generalized tensor factorization, which is flexible enough to choose from a variate of losses to best suit different types of data in practice. We design a three-level communication reduction strategy tailored to the generalized tensor factorization, which is able to reduce the uplink communication cost up to 99.90%. In addition, we theoretically prove that our algorithm does not compromise convergence speed despite the aggressive communication compression. Extensive experiments on two real-world electronics health record datasets demonstrate the efficiency improvements in terms of computation and communication cost.
Lexical phenomena, such as clusters of words, disseminate through social networks at different rates but most models of diffusion focus on the discrete adoption of new lexical phenomena (i.e. new topics or memes). It is possible much of lexical diffusion happens via the changing rates of existing word categories or concepts (those that are already being used, at least to some extent, regularly) rather than new ones. In this study we introduce a new metric, contrastive lexical diffusion (CLD) coefficient, which attempts to measure the degree to which ordinary language (here clusters of common words) catch on over friendship connections over time. For instance topics related to meeting and job are found to be sticky, while negative thinking and emotion, and global events, like ‘school orientation’ were found to be less sticky even though they change rates over time. We evaluate CLD coefficient over both quantitative and qualitative tests, studied over 6 years of language on Twitter. We find CLD predicts the spread of tweets and friendship connections, scores converge with human judgments of lexical diffusion (r=0.92), and CLD coefficients replicate across disjoint networks (r=0.85). Comparing CLD scores can help understand lexical diffusion: positive emotion words appear more diffusive than negative emotions, first-person plurals (we) score higher than other pronouns, and numbers and time appear non-contagious.
Research suggests that social relationships have substantial impacts on individuals' health outcomes. Network intervention, through careful planning, can assist a network of users to build healthy relationships. However, most previous work is not designed to assist such planning by carefully examining and improving multiple network characteristics. In this paper, we propose and evaluate algorithms that facilitate network intervention planning through simultaneous optimization of network degree, closeness, betweenness, and local clustering coefficient, under scenarios involving Network Intervention with Limited Degradation - for Single target (NILD-S) and Network Intervention with Limited Degradation - for Multiple targets (NILD-M). We prove that NILD-S and NILD-M are NP-hard and cannot be approximated within any ratio in polynomial time unless P=NP. We propose the Candidate Re-selection with Preserved Dependency (CRPD) algorithm for NILD-S, and the Objective-aware Intervention edge Selection and Adjustment (OISA) algorithm for NILD-M. Various pruning strategies are designed to boost the efficiency of the proposed algorithms. Extensive experiments on various real social networks collected from public schools and Web and an empirical study are conducted to show that CRPD and OISA outperform the baselines in both efficiency and effectiveness.
Predicting pregnancy has been a fundamental problem in women's health for more than 50 years. Previous datasets have been collected via carefully curated medical studies, but the recent growth of women's health tracking mobile apps offers potential for reaching a much broader population. However, the feasibility of predicting pregnancy from mobile health tracking data is unclear. Here we develop four models -- a logistic regression model, and 3 LSTM models -- to predict a woman's probability of becoming pregnant using data from a women's health tracking app, Clue by BioWink GmbH. Evaluating our models on a dataset of 79 million logs from 65,276 women with ground truth pregnancy test data, we show that our predicted pregnancy probabilities meaningfully stratify women: women in the top 10% of predicted probabilities have a 89% chance of becoming pregnant over 6 menstrual cycles, as compared to a 27% chance for women in the bottom 10%. We develop a technique for extracting interpretable time trends from our deep learning models, and show these trends are consistent with previous fertility research. Our findings illustrate the potential that women's health tracking data offers for predicting pregnancy on a broader population; we conclude by discussing the steps needed to fulfill this potential.
Activity tracking apps often make use of goals as one of their core motivational tools. There are two critical components to this tool: setting a goal, and subsequently achieving that goal. Despite its crucial role in how a number of prominent self-tracking apps function, there has been relatively little investigation of the goal-setting and achievement aspects of self-tracking apps. Here we explore this issue, investigating a particular goal setting and achievement process that is extensive, recorded, and crucial for both the app and its users' success: weight loss goals in MyFitnessPal. We present a large-scale study of 1.4 million users and weight loss goals, allowing for an unprecedented detailed view of how people set and achieve their goals. We find that, even for difficult long-term goals, behavior within the first 7 days predicts those who ultimately achieve their goals, that is, those who lose at least as much weight as they set out to, and those who do not. For instance, high amounts of early weight loss, which some researchers have classified as unsustainable, leads to higher goal achievement rates. We also show that early food intake, self-monitoring motivation, and attitude towards the goal are important factors. We then show that we can use our findings to predict goal achievement with an accuracy of 79% ROC AUC just 7 days after a goal is set. Finally, we discuss how our findings could inform steps to improve goal achievement in self-tracking apps.
In the past few decades, there has been rapid growth in quantity and variety of healthcare data. These large sets of data are usually high dimensional (e.g. patients, their diagnoses, and medications to treat their diagnoses) and cannot be adequately represented as matrices. Thus, many existing algorithms can not analyze them. To accommodate these high dimensional data, tensor factorization, which can be viewed as a higher-order extension of methods like PCA, has attracted much attention and emerged as a promising solution. However, tensor factorization is a computationally expensive task, and existing methods developed to factor large tensors are not flexible enough for real-world situations. To address this scaling problem more efficiently, we introduce SGranite, a distributed, scalable, and sparse tensor factorization method fit through stochastic gradient descent. SGranite offers three contributions: (1) Scalability: it employs a block partitioning and parallel processing design and thus scales to large tensors, (2) Accuracy: we show that our method can achieve results faster without sacrificing the quality of the tensor decomposition, and (3) FlexibleConstraints: we show our approach can encompass various kinds of constraints including l2 norm, l1 norm, and logistic regularization. We demonstrate SGranite's capabilities in two real-world use cases. In the first, we use Google searches for flu-like symptoms to characterize and predict influenza patterns. In the second, we use SGranite to extract clinically interesting sets (i.e., phenotypes) of patients from electronic health records. Through these case studies, we show SGranite has the potential to be used to rapidly characterize, predict, and manage a large multimodal datasets, thereby promising a novel, data-driven solution that can benefit very large segments of the population.
Mobile health applications, including those that track activities such as exercise, sleep, and diet, are becoming widely used. Accurately predicting human actions in the real world is essential for targeted recommendations that could improve our health and for personalization of these applications. However, making such predictions is extremely difficult due to the complexities of human behavior, which consists of a large number of potential actions that vary over time, depend on each other, and are periodic. Previous work has not jointly modeled these dynamics and has largely focused on item consumption patterns instead of broader types of behaviors such as eating, commuting or exercising. In this work, we develop a novel statistical model, called TIPAS, for Time-varying, Interdependent, and Periodic Action Sequences. Our approach is based on personalized, multivariate temporal point processes that model time-varying action propensities through a mixture of Gaussian intensities. Our model captures short-term and long-term periodic interdependencies between actions through Hawkes process-based self-excitations. We evaluate our approach on two activity logging datasets comprising 12 million real-world actions (e.g., eating, sleep, and exercise) taken by 20 thousand users over 17 months. We demonstrate that our approach allows us to make successful predictions of future user actions and their timing. Specifically, TIPAS improves predictions of actions, and their timing, over existing methods across multiple datasets by up to 156%, and up to 37%, respectively. Performance improvements are particularly large for relatively rare and periodic actions such as walking and biking, improving over baselines by up to 256%. This demonstrates that explicit modeling of dependencies and periodicities in real-world behavior enables successful predictions of future actions, with implications for modeling human behavior, app personalization, and targeting of health interventions.
Many problems in areas as diverse as recommendation systems, social network analysis, semantic search, and distributed root cause analysis can be modeled as pattern search on labeled graphs (also called "heterogeneous information networks" or HINs). Given a large graph and a query pattern with node and edge label constraints, a fundamental challenge is to find the top-k matches according to a ranking function over edge and node weights. For users, it is difficult to select value k. We therefore propose the novel notion of an any-k ranking algorithm: for a given time budget, return as many of the top-ranked results as possible. Then, given additional time, produce the next lower-ranked results quickly as well. It can be stopped anytime, but may have to continue until all results are returned. This paper focuses on acyclic patterns over arbitrary labeled graphs. We are interested in practical algorithms that effectively exploit (1) properties of heterogeneous networks, in particular selective constraints on labels, and (2) that the users often explore only a fraction of the top-ranked results. Our solution, KARPET, carefully integrates aggressive pruning that leverages the acyclic nature of the query, and incremental guided search. It enables us to prove strong non-trivial time and space guarantees, which is generally considered very hard for this type of graph search problem. Through experimental studies we show that KARPET achieves running times in the order of milliseconds for tree patterns on large networks with millions of nodes and edges.
Cycles are fundamental to human health and behavior. Examples include mood cycles, circadian rhythms, and the menstrual cycle. However, modeling cycles in time series data is challenging because in most cases the cycles are not labeled or directly observed and need to be inferred from multidimensional measurements taken over time. Here, we present Cyclic Hidden Markov Models (CyHMMs) for detecting and modeling cycles in a collection of multidimensional heterogeneous time series data. In contrast to previous cycle modeling methods, CyHMMs deal with a number of challenges encountered in modeling real-world cycles: they can model multivariate data with both discrete and continuous dimensions; they explicitly model and are robust to missing data; and they can share information across individuals to accommodate variation both within and between individual time series. Experiments on synthetic and real-world health-tracking data demonstrate that CyHMMs infer cycle lengths more accurately than existing methods, with 58% lower error on simulated data and 63% lower error on real-world data compared to the best-performing baseline. CyHMMs can also perform functions which baselines cannot: they can model the progression of individual features/symptoms over the course of the cycle, identify the most variable features, and cluster individual time series into groups with distinct characteristics. Applying CyHMMs to two real-world health-tracking datasets -- of human menstrual cycle symptoms and physical activity tracking data -- yields important insights including which symptoms to expect at each point during the cycle. We also find that people fall into several groups with distinct cycle patterns, and that these groups differ along dimensions not provided to the model. For example, by modeling missing data in the menstrual cycles dataset, we are able to discover a medically relevant group of birth control users even though information on birth control is not given to the model.
Mobile health applications that track activities, such as exercise, sleep, and diet, are becoming widely used. While these activity tracking applications have the potential to improve our health, user engagement and retention are critical factors for their success. However, long-term user engagement patterns in real-world activity tracking applications are not yet well understood. Here we study user engagement patterns within a mobile physical activity tracking application consisting of 115 million logged activities taken by over a million users over 31 months. Specifically, we show that over 75% of users return and re-engage with the application after prolonged periods of inactivity, no matter the duration of the inactivity. We find a surprising result that the re-engagement usage patterns resemble those of the start of the initial engagement period, rather than being a simple continuation of the end of the initial engagement period. This evidence points to a conceptual model of multiple lives of user engagement, extending the prevalent single life view of user activity. We demonstrate that these multiple lives occur because the users have a variety of different primary intents or goals for using the app. These primary intents are associated with how long each life lasts and how likely the user is to re-engage for a new life. We find evidence for users being more likely to stop using the app once they achieved their primary intent or goal (e.g., weight loss). However, these users might return once their original intent resurfaces (e.g., wanting to lose newly gained weight). We discuss implications of the multiple life paradigm and propose a novel prediction task of predicting the number of lives of a user. Based on insights developed in this work, including a marker of improved primary intent performance, our prediction models achieve 71% ROC AUC. Overall, our research has implications for modeling user re-engagement in health activity tracking applications and has consequences for how notifications, recommendations as well as gamification can be used to increase engagement.
