This is an old revision of the document!


Research

The Vis Lab is focused on the application of artificial intelligence and machine learning to problems in the Learning Health System (LHS) that include:

  • Learning electronic medical record (EMR) system and computerized clinical decision support
  • Precision medicine and personalized modeling
  • Data mining and causal discovery from biomedical data
  • Reuse of EMR data and research data warehousing
  • Automated visual analytics

The goal of a LHS is to deliver the best care every time, and to learn and improve with each care experience. A LHS has two arms: 1) an afferent (blue) arm that is focused on assembling data from various sources including EMR systems, mobile health (mHealth) applications and research studies into an integrated research data repository, and 2) an efferent (red) arm that is focused on returning results and findings obtained from analyses of the data repository to inform clinical decision support and patient decision support systems.

Learning electronic medical record (EMR) system and computerized clinical decision support

This work is funded by a R01 grant from NLM, NIH.

Electronic medical records (EMRs) are capturing increasing amounts of patient data that can be leveraged by machine learning methods for computerized clinical decision support. My work focuses on developing a learning EMR system that uses machine learning to provide decision support using the right data, at the right time. In addition, I work with a team of collaborators in developing and implementing machine learning methods for detecting adverse drug events and for identifying anomalies in clinical management of patients. This work is in collaboration with Gregory F. Cooper, Harry Hochheiser, Milos Hauskrecht, Gilles Clermont, (at the University of Pittsburgh), and Dean Sittig (at the University of Texas Health Science Center at Houston).

Publications:

  • Calzoni, L, Clermont, G, Cooper, GF, Visweswaran, S, Hochheiser, H. Exploring novel graphical representations of clinical data in a learning EMR. In: AMIA Annual Symposium Proceedings. 2017 Nov 7. pdf
  • King, AJ, Hochheiser, H, Visweswaran, S, Clermont, G, Cooper, GF. Eye-tracking for clinical decision support: A method to capture automatically what physicians are viewing in the EMR. In: AMIA Joint Summits Translational Science Proceedings. 2017 Mar 27-30; 2017:512-21. PMID: 28815151 PMCID: PMC5543363 pdf (Awarded First Place in the Student Paper Competition at the AMIA Joint Summits Clinical Research Informatics, 2017.)
  • King, AJ, Cooper, GF, Hochheiser, H, Clermont, G, Visweswaran, S. Development and preliminary evaluation of a prototype of a learning electronic medical record system. In: AMIA Annual Symposium Proceedings. 2015 Nov 17; 2015:1967-75. PMID: 26958296 PMCID: PMC4765593 pdf (Awarded First Place in the Student Paper Competition at the AMIA Annual Symposium, 2015.)

Precision medicine and patient-specific modeling

This work is funded by a UG3 grant from the Office of the Director, NIH.

The University of Pittsburgh is funded as one of the Healthcare Provider Organizations for the national All of Us Research Program, which is a historic effort to gather data from 1 million people living in the United States. The goal of the program is to revolutionize how disease is prevented and treated based on individual differences in lifestyle, environment and genetics. Along with Steven E. Reis and Oscar Marroquin, I am a PD/PI on the All of Us Pennsylvania project that will enroll and collect data and bio specimens for 120,000 individuals for the All of Us Research Program.

In predictive modeling in medicine, the typical paradigm consists of learning a single model from a database of individuals, which is then applied to predict outcomes for any future individual. Such a model is called a population-wide model because it is intended to be applied to an entire population of future individuals. In contrast, patient-specific modeling focuses on learning models that are tailored to the characteristics of the individual at hand. Patient-specific models that are optimized to perform well for a specific individual are likely to have better predictive performance than the typical population-wide models that are optimized to have good predictive performance on average on all future individuals. Moreover, patient-specific models can identify features such as genomic factors that are specific for an individual thus enabling precision medicine. This work is in collaboration with Gregory F. Cooper (at the University of Pittsburgh). The areas of focus include:

  • development of Bayesian and information-theoretic methods for learning patient-specific models from clinical and genomic data, and
  • application of personalized modeling to risk assessment, diagnosis, prognosis and selection of therapy.

Publications:

Data mining and causal discovery from biomedical data

Development of efficient data mining for biomarker discovery from high-dimensional genomic variant data is important for precision medicine and focuses on the use of information obtained from sequences such as whole exomes and whole genomes. Our work focuses on single nucleotide variants (SNVs) data obtained from genome-wide and whole exome data. This work is in collaboration with Gregory F. Cooper and David C. Whitcomb (at the University of Pittsburgh).

Development and application of causal discovery machine learning methods to biomedical data including large-scale electronic medical record data is important for uncovering casual relations in these data. This work is in collaboration with Gregory F. Cooper and Eric Strobl and is funded by a U54 grant from NHGRI, NIH.

Surging use of social media has resulted in unprecedented opportunities for examination of complex social and behavioral phenomena. We are developing machine learning infoveillance tools to study cancer communication in Twitter data. This work is in collaboration with This work is in collaboration with Brian Primack and is funded by a R01 grant from NCI, NIH.

The areas of focus include:

  • development of predictive models from high-dimensional genomic data,
  • development of algorithms to infer causal relationships from observational data, and
  • application of machine learning to build an infoveillance tool to study cancer communication in Twitter data.

Publications:

  • Strobl, EV, Visweswaran, S. Deep multiple kernel learning. In: Proceedings of the 12th International Conference on Machine Learning and Applications (ICMLA'13). 2013 Dec 4; 2013:414-17.

Reuse of EMR data and research data warehousing

This work is funded by a UL1 grant from NCATS, NIH and a CDRN grant from PCORI.

Several local, regional and national efforts are ongoing that are creating clinical data repositories for reuse of electronic medical record data for clinical, translational, and informatics research. I lead the development and implementation of a Pitt research data warehouse called Neptune. I also lead the efforts for data harmonization, translation to standard terminologies and mapping to standard value sets for several projects that include: 1) NCATS-funded Accrual of patients to Clinical Trials (ACT) network, 2) NIH-funded All of Us Research Program, and 3) PCORI-funded PaTH clinical data research network (CDRN). This work is in collaboration with Steven E. Reis, Michael J. Becich, and Melissa Saul (at the University of Pittsburgh).

Publications:

  • Amin, W, Borromeo, C, Saul, M, Becich, MJ, Visweswaran, S. Informatics synergies between PaTH and ACT networks. In: Proceedings of the 2015 Summit on Clinical Research Informatics (Mar 2015).
  • Visweswaran, S, Tenenbaum, J, Gouripeddi, R. Secondary use of data for research - EHR, omics and environmental data. In: AMIA Joint Summits Translational Science Proceedings (Mar 2016).

Automated visual analytics

Automated visual analytics combines visual analytics with automated analysis for discovery of patterns in data. Visual analytics is interactive analysis facilitated by interactive visual interfaces where the domain expert interacts with the data visually to identify interesting patterns. However, it is effort-intensive and is readily applicable only to datasets with low dimensionality and small sample sizes. Our work focusses on automated visual analytics that combines visualization and automated analysis methods to take advantage of the rapid search that automated methods provide with the ability to identify novel and rich patterns that visual analytics provides. This work is in collaboration with Suresh K Bhavnani (at the University of Texas Medical Branch at Galveston), Kevin Bassler (University of Houston, Central), and Rohit D Divaker (Mayo Clinic, Rochester). The areas of focus include:

  • development of automated visual analytics methods that combine bipartite network visual analytics with heuristic search methods from machine-learning, and
  • application of automated visual analytics methods to genomic data for discovery of gene-gene interactions and for discovery of sub-phenotypes.

Publications: