Advanced Computational Framework for Decision Support in Critically Ill Children This application addresses broad Challenge Area (10): Increasing Technology for Processing Health Care Data and specific Challenge Topic 10-LM-102: Advanced decision support for complex clinical decisions. Abstract: Artificial Intelligence (AI) and advanced computational techniques, applied to complex, multidimensional, streaming, clinical data (historic, physiologic, laboratory, imaging, etc) from disparate health care digital data sources, can be used to produce integrated, higher level representations of critically ill patients for knowledge discovery and decision support for the 'next'patient. Vast amounts of digital health care data are available for real time computational analysis using artificial intelligence and statistical approaches such as data clustering and neural networks to ascertain relationships, determine 'diagnostic clusters'and trend outcomes of therapy in individual and groups of patients. While such algorithms have successful applications in business, industry, physical sciences and in parts of health care, several barriers exist for their broader application to intensive care medicine and other health care domains. Using AI and automated computational methodology we will develop algorithms for data mining raw medical data from disparate clinical data sources. This will enable understanding and application of the most recent experiential information from large numbers of critically ill patients to find similarities between a current individual patient and historical, similar populations with known treatment outcomes. This will iteratively lead to a refined representation of the individual patient and guide patient management. This project is not merely the development of a database, although this is an essential precursor for the application of the advanced analytic techniques we will develop. It is ultimately about developing a framework to support the extraction, manipulation and construction of new views, relationships and information of and from observational clinical data from multiple sources to enable meaningful analysis and application at the patient's bedside. There are many open questions not sufficiently addressed about the translation of raw clinical data into an analytic computational infrastructure capable of providing bedside decision support. We will integrate expertise in medicine, computational mathematics and computer science to construct suitable data structures and develop AI computational techniques for analysis and presentation. Our goal is to advance the development of explicit methodology to bring analyzed comparative data to the bedside of critically ill children in real time to support clinical decision-making. Such a process will involve: 1.) transformation of patient-centric data across disparate platforms into a large relational database;2.) construction of an integrated data model, assuring data integrity, structure and security;and finally, 3.) data analysis and presentation as meaningful comparative metrics that adequately characterize the time-based continuum of critical illness. Analytics can then be developed and algorithms refined that can be applied to severity of illness, prognostic, therapeutic, and anomaly detection in patients, hospital units, and national populations, thus painting a holistic picture of patients and populations. Our goal is to provide an integrated high-level view of a patient compared to and in the context of previous critically ill patients. We will use advanced computational techniques and artificial intelligence to detect categories within raw medical data from disparate data sources allowing the most recent experiential information about large numbers of critically ill children to be mined to find similarities between the index case and historical cases with known outcomes. This will enable decision support for diagnosis, management, therapy and outcomes serving the functions of disease detection, direct patient care, quality, and safety.
Our goal is to provide an integrated high-level view of a patient compared to and in the context of previous critically ill patients. We will use advanced computational techniques and artificial intelligence to detect categories within raw medical data from disparate data sources allowing the most recent experiential information about large numbers of critically ill children to be mined to find similarities between the index case and historical cases with known outcomes. This will enable decision support for diagnosis, management, therapy and outcomes serving the functions of disease detection, direct patient care, quality, and safety.
| Khemani, Robinder G; Thomas, Neal J; Venkatachalam, Vani et al. (2012) Comparison of SpO2 to PaO2 based markers of lung disease severity for children with acute lung injury. Crit Care Med 40:1309-16 |
| Ghuman, Anoopindar K; Newth, Christopher J L; Khemani, Robinder G (2012) The association between the end tidal alveolar dead space fraction and mortality in pediatric acute hypoxemic respiratory failure. Pediatr Crit Care Med 13:11-5 |
| Khemani, Robinder G; Newth, Christopher J L (2010) The design of future pediatric mechanical ventilation trials for acute lung injury. Am J Respir Crit Care Med 182:1465-74 |
