Machine Learning and Clinical Text. Supporting Health Information Flow

Fluent health information flow is critical for clinical decision-making. However, a considerable part of this information is free-form text and inabilities to utilize it create risks to patient safety and cost-­effective hospital administration. Methods for automated processing of clinical text are emerging. The aim in this doctoral dissertation is to study machine learning and clinical text in order to support health information flow.First, by analyzing the content of authentic patient records, the aim is to specify clinical needs in order to guide the development of machine learning applications.The contributions are a model of the ideal information flow,a model of the problems and challenges in reality, and a road map for the technology development. Second, by developing applications for practical cases,the aim is to concretize ways to support health information flow. Altogether five machine learning applications for three practical cases are described: The first two applications are binary classification and regression related to the practical case of topic labeling and relevance ranking.The third and fourth application are supervised and unsupervised multi-class classification for the practical case of topic segmentation and labeling.These four applications are tested with Finnish intensive care patient records.The fifth application is multi-label classification for the practical task of diagnosis coding. It is tested with English radiology reports.The performance of all these applications is promising. Third, the aim is to study how the quality of machine learning applications can be reliably evaluated.The associations between performance evaluation measures and methods are addressed,and a new hold-out method is introduced.This method contributes not only to processing time but also to the evaluation diversity and quality. The main conclusion is that developing machine learning applications for text requires interdisciplinary, international collaboration. Practical cases are very different, and hence the development must begin from genuine user needs and domain expertise. The technological expertise must cover linguistics,machine learning, and information systems. Finally, the methods must be evaluated both statistically and through authentic user-feedback.

Capturing temporality and intentionality in information systems

Information modelling is a topic that has been researched a great deal, but still many questions around it have not been solved. An information model is essential in the design of a database which is the core of an information system. Currently most of databases only deal with information that represents facts, or asserted information. The ability of capturing semantic aspect has to be improved, and yet other types, such as temporal and intentional information, should be considered. Semantic Analysis, a method of information modelling, has offered a way to handle various aspects of information. It employs the domain knowledge and communication acts as sources of information modelling. It lends itself to a uniform structure whereby semantic, temporal and intentional information can be captured, which builds a sound foundation for building a semantic temporal database.

DEVELOPING A METHOD FOR IDENTIFYING AND REDUCING FUNCTIONAL DISCREPANCIES OF INFORMATION SYSTEMS

Currently more than half of Electronic Health Record (EHR) projects fail. Most of these failures are not due to flawed technology, but rather due to the lack of systematic considerations of human issues. Among the barriers for EHR adoption, function mismatching among users, activities, and systems is a major area that has not been systematically addressed from a human-centered perspective. A theoretical framework called Functional Framework was developed for identifying and reducing functional discrepancies among users, activities, and systems. The Functional Framework is composed of three models – the User Model, the Designer Model, and the Activity Model. The User Model was developed by conducting a survey (N = 32) that identified the functions needed and desired from the user’s perspective. The Designer Model was developed by conducting a systemic review of an Electronic Dental Record (EDR) and its functions. The Activity Model was developed using an ethnographic method called shadowing where EDR users (5 dentists, 5 dental assistants, 5 administrative personnel) were followed quietly and observed for their activities. These three models were combined to form a unified model. From the unified model the work domain ontology was developed by asking users to rate the functions (a total of 190 functions) in the unified model along the dimensions of frequency and criticality in a survey. The functional discrepancies, as indicated by the regions of the Venn diagrams formed by the three models, were consistent with the survey results, especially with user satisfaction. The survey for the Functional Framework indicated the preference of one system over the other (R=0.895). The results of this project showed that the Functional Framework provides a systematic method for identifying, evaluating, and reducing functional discrepancies among users, systems, and activities. Limitations and generalizability of the Functional Framework were discussed.

Toxics Release Inventory facilities and childhood cancer: Geographic information systems based approach

Purpose. To examine the association between living in proximity to Toxics Release Inventory (TRI) facilities and the incidence of childhood cancer in the State of Texas. ^ Design. This is a secondary data analysis utilizing the publicly available Toxics release inventory (TRI), maintained by the U.S. Environmental protection agency that lists the facilities that release any of the 650 TRI chemicals. Total childhood cancer cases and childhood cancer rate (age 0-14 years) by county, for the years 1995-2003 were used from the Texas cancer registry, available at the Texas department of State Health Services website. Setting: This study was limited to the children population of the State of Texas. ^ Method. Analysis was done using Stata version 9 and SPSS version 15.0. Satscan was used for geographical spatial clustering of childhood cancer cases based on county centroids using the Poisson clustering algorithm which adjusts for population density. Pictorial maps were created using MapInfo professional version 8.0. ^ Results. One hundred and twenty five counties had no TRI facilities in their region, while 129 facilities had at least one TRI facility. An increasing trend for number of facilities and total disposal was observed except for the highest category based on cancer rate quartiles. Linear regression analysis using log transformation for number of facilities and total disposal in predicting cancer rates was computed, however both these variables were not found to be significant predictors. Seven significant geographical spatial clusters of counties for high childhood cancer rates (p<0.05) were indicated. Binomial logistic regression by categorizing the cancer rate in to two groups (<=150 and >150) indicated an odds ratio of 1.58 (CI 1.127, 2.222) for the natural log of number of facilities. ^ Conclusion. We have used a unique methodology by combining GIS and spatial clustering techniques with existing statistical approaches in examining the association between living in proximity to TRI facilities and the incidence of childhood cancer in the State of Texas. Although a concrete association was not indicated, further studies are required examining specific TRI chemicals. Use of this information can enable the researchers and public to identify potential concerns, gain a better understanding of potential risks, and work with industry and government to reduce toxic chemical use, disposal or other releases and the risks associated with them. TRI data, in conjunction with other information, can be used as a starting point in evaluating exposures and risks. ^

Health information exchange: Using metrics to address quality of care and return on investment

Health Information Exchange (HIE) will play a key part in our nation’s effort to improve healthcare. The evidence of HIEs transformational role in healthcare delivery systems is quite limited. The lack of such evidence led us to explore what exists in the healthcare industry that may provide evidence of effectiveness and efficiency of HIEs. The objective of the study was to find out how many fully functional HIEs are using any measurements or metrics to gauge impact of HIE on quality improvement (QI) and on return on investment (ROI).^ A web-based survey was used to determine the number of operational HIEs using metrics for QI and ROI. Our study highlights the fact that only 50 percent of the HIEs who responded use or plan to use metrics. However, 95 percent of the respondents believed HIEs improve quality of care while only 56 percent believed HIE showed positive ROI. Although operational HIEs present numerous opportunities to demonstrate the business model for improving health care quality, evidence to document the impact of HIEs is lacking. ^

Endemic tuberculosis, the homeless, and public transportation: A merging of geographical information systems surveillance and the Houston Tuberculosis Initiative Surveillance project

To reach the goals established by the Institute of Medicine (IOM) and the Centers for Disease Control's (CDC) STOP TB USA, measures must be taken to curtail a future peak in Tuberculosis (TB) incidence and speed the currently stagnant rate of TB elimination. Both efforts will require, at minimum, the consideration and understanding of the third dimension of TB transmission: the location-based spread of an airborne pathogen among persons known and unknown to each other. This consideration will require an elucidation of the areas within the U.S. that have endemic TB. The Houston Tuberculosis Initiative (HTI) was a population-based active surveillance of confirmed Houston/Harris County TB cases from 1995–2004. Strengths in this dataset include the molecular characterization of laboratory confirmed cases, the collection of geographic locations (including home addresses) frequented by cases, and the HTI time period that parallels a decline in TB incidence in the United States (U.S.). The HTI dataset was used in this secondary data analysis to implement a GIS analysis of TB cases, the locations frequented by cases, and their association with risk factors associated with TB transmission. ^ This study reports, for the first time, the incidence of TB among the homeless in Houston, Texas. The homeless are an at-risk population for TB disease, yet they are also a population whose TB incidence has been unknown and unreported due to their non-enumeration. The first section of this dissertation identifies local areas in Houston with endemic TB disease. Many Houston TB cases who reported living in these endemic areas also share the TB risk factor of current or recent homelessness. Merging the 2004–2005 Houston enumeration of the homeless with historical HTI surveillance data of TB cases in Houston enabled this first-time report of TB risk among the homeless in Houston. The homeless were more likely to be US-born, belong to a genotypic cluster, and belong to a cluster of a larger size. The calculated average incidence among homeless persons was 411/100,000, compared to 9.5/100,000 among housed. These alarming rates are not driven by a co-infection but by social determinants. The unsheltered persons were hospitalized more days and required more follow-up time by staff than those who reported a steady housing situation. The homeless are a specific example of the increased targeting of prevention dollars that could occur if TB rates were reported for specific areas with known health disparities rather than as a generalized rate normalized over a diverse population. ^ It has been estimated that 27% of Houstonians use public transportation. The city layout allows bus routes to run like veins connecting even the most diverse of populations within the metropolitan area. Secondary data analysis of frequent bus use (defined as riding a route weekly) among TB cases was assessed for its relationship with known TB risk factors. The spatial distribution of genotypic clusters associated with bus use was assessed, along with the reported routes and epidemiologic-links among cases belonging to the identified clusters. ^ TB cases who reported frequent bus use were more likely to have demographic and social risk factors associated with poverty, immune suppression and health disparities. An equal proportion of bus riders and non-bus riders were cultured for Mycobacterium tuberculosis, yet 75% of bus riders were genotypically clustered, indicating recent transmission, compared to 56% of non-bus riders (OR=2.4, 95%CI(2.0, 2.8), p<0.001). Bus riders had a mean cluster size of 50.14 vs. 28.9 (p<0.001). Second order spatial analysis of clustered fingerprint 2 (n=122), a Beijing family cluster, revealed geographic clustering among cases based on their report of bus use. Univariate and multivariate analysis of routes reported by cases belonging to these clusters found that 10 of the 14 clusters were associated with use. Individual Metro routes, including one route servicing the local hospitals, were found to be risk factors for belonging to a cluster shown to be endemic in Houston. The routes themselves geographically connect the census tracts previously identified as having endemic TB. 78% (15/23) of Houston Metro routes investigated had one or more print groups reporting frequent use for every HTI study year. We present data on three specific but clonally related print groups and show that bus-use is clustered in time by route and is the only known link between cases in one of the three prints: print 22. (Abstract shortened by UMI.)^

Correlation of tuberculosis and human immunodeficiency virus in Harris County, Texas from 2009 through 2010 using Geographic Information Systems

A population based ecological study was conducted to identify areas with a high number of TB and HIV new diagnoses in Harris County, Texas from 2009 through 2010 by applying Geographic Information Systems to determine whether distinguished spatial patterns exist at the census tract level through the use of exploratory mapping. As of 2010, Texas has the fourth highest occurrence of new diagnoses of HIV/AIDS and TB.[31] The Texas Department of State Health Services (DSHS) has identified HIV infected persons as a high risk population for TB in Harris County.[29] In order to explore this relationship further, GIS was utilized to identify spatial trends. ^ The specific aims were to map TB and HIV new diagnoses rates and spatially identify hotspots and high value clusters at the census tract level. The potential association between HIV and TB was analyzed using spatial autocorrelation and linear regression analysis. The spatial statistics used were ArcGIS 9.3 Hotspot Analysis and Cluster and Outlier Analysis. Spatial autocorrelation was determined through Global Moran's I and linear regression analysis. ^ Hotspots and clusters of TB and HIV are located within the same spatial areas of Harris County. The areas with high value clusters and hotspots for each infection are located within the central downtown area of the city of Houston. There is an additional hotspot area of TB located directly north of I-10 and a hotspot area of HIV northeast of Interstate 610. ^ The Moran's I Index of 0.17 (Z score = 3.6 standard deviations, p-value = 0.01) suggests that TB is statistically clustered with a less than 1% chance that this pattern is due to random chance. However, there were a high number of features with no neighbors which may invalidate the statistical properties of the test. Linear regression analysis indicated that HIV new diagnoses rates (β=−0.006, SE=0.147, p=0.970) and census tracts (β=0.000, SE=0.000, p=0.866) were not significant predictors of TB new diagnoses rates. ^ Mapping products indicate that census tracts with overlapping hotspots and high value clusters of TB and HIV should be a targeted focus for prevention efforts, most particularly within central Harris County. While the statistical association was not confirmed, evidence suggests that there is a relationship between HIV and TB within this two year period.^

The information systems integration for a better caring and chronic patients´ empowerment

The technology innovation (TI) in the health system has been confirmed as the best format to share information and to facilitate the communication between all the actors involved in the chronic patients´ care.

Emergency medical service systems and health planning : a bibliography.

Mode of access: Internet.

Information systems design for the community health services

This system is concerned with the design and implementation of a community health information system which fulfils some of the local needs of fourteen nursing and para-medical professions in a district health authority, whilst satisfying the statutory requirements of the NHS Korner steering group for those professions. A national survey of community health computer applications, documented in the form of an applications register, shows the need for such a system. A series of general requirements for an informations systems design methodology are identified, together with specific requirements for this problem situation. A number of existing methodologies are reviewed, but none of these were appropriate for this application. Some existing approaches, tools and techniques are used to define a more suitable methodology. It is unreasonable to rely on one single general methodology for all types of application development. There is a need for pragmatism, adaptation and flexibility. In this research, participation in the development stages by those who will eventually use the system was thought desirable. This was achieved by forming a representative design group. Results would seem to show a highly favourable response from users to this participation which contributed to the overall success of the system implemented. A prototype was developed for the chiropody and school nursing staff groups of Darlington health authority, and evaluations show that a significant number of the problems and objectives of those groups have been successfully addressed; the value of community health information has been increased; and information has been successfully fed back to staff and better utilised.

The evaluation of information systems in the organisational context of the National Health Service. Available in 2 volumes.

This thesis describes a project which has investigated the evaluation of information systems. The work took place in, and is related to, a specific organisational context, that of the National Health Service (NHS). It aims to increase understanding of the evaluation which takes place in the service and the way in which this is affected by the NHS environment. It also investigates the issues which surround some important types of evaluation and their use in this context. The first stage of the project was a postal survey in which respondents were asked to describe the evaluation which took place in their authorities and to give their opinions about it. This was used to give an overview of the practice of IS evaluation in the NHS and to identify its uses and the problems experienced. Three important types of evaluation were then examined in more detail by means of action research studies. One of these dealt with the selection and purchase of a large hospital information system. The study took the form of an evaluation of the procurement process, and examined the methods used and the influence of organisational factors. The other studies are concerned with post-implementation evaluation, and examine the choice of an evaluation approach as well as its application. One was an evaluation of a community health system which had been operational for some time but was of doubtful value, and suffered from a number of problems. The situation was explored by means of a study of the costs and benefits of the system. The remaining study was the initial review of a system which was used in the administration of a Breast Screening Service. The service itself was also newly operational and the relationship between the service and the system was of interest.

The Evaluation of health and safety auditing systems

The specific objective of the research was to evaluate proprietary audit systems. Proprietary audit systems comprise question sets containing approximately 500 questions dealing with selected aspects of health and safety management. Each question is allotted a number of points and an organisation seeks to judge its health and safety performance by the overall score achieved in the audit. Initially it was considered that the evaluation method might involve comparing the proprietary audit scores with other methods of measuring safety performance. However, what appeared to be missing in the first instance was information that organisations could use to compare the contrast question set content against their own needs. A technique was developed using the computer database FileMaker Pro. This enables questions in an audit to be sorted into categories using a process of searching for key words. Questions that are not categorised by word searching can be identified and sorted manually. The process can be completed in 2-3 hours which is considerably faster than manual categorisation of questions which typically takes about 10 days. The technique was used to compare and contrast three proprietary audits: ISRS, CHASE and QSA. Differences and similarities between these audits were successfully identified. It was concluded that in general proprietary audits need to focus to a greater extent on identifying strengths and weaknesses in occupational health and safety management systems. To do this requires the inclusion of more probing questions which consider whether risk control measures are likely to be successful.

The notion of structure and the information records of the documentary systems

Assuming as a starting point the acknowledge that the principles and methods used to build and manage the documentary systems are disperse and lack systematization, this study hypothesizes that the notion of structure, when assuming mutual relationships among its elements, promotes more organical systems and assures better quality and consistency in the retrieval of information concerning users` matters. Accordingly, it aims to explore the fundamentals about the records of information and documentary systems, starting from the notion of structure. In order to achieve that, it presents basic concepts and relative matters to documentary systems and information records. Next to this, it lists the theoretical subsides over the notion of structure, studied by Benveniste, Ferrater Mora, Levi-Strauss, Lopes, Penalver Simo, Saussure, apart from Ducrot, Favero and Koch. Appropriations that have already been done by Paul Otlet, Garcia Gutierrez and Moreiro Gonzalez. In Documentation come as a further topic. It concludes that the adopted notion of structure to make explicit a hypothesis of real systematization achieves more organical systems, as well as it grants pedagogical reference to the documentary tasks.