Predicting 30-day Hospital Readmission with Publicly Available Administrative Database. A Conditional Logistic Regression Modeling Approach.

INTRODUCTION: This article is part of the Focus Theme of Methods of Information in Medicine on "Big Data and Analytics in Healthcare". BACKGROUND: Hospital readmissions raise healthcare costs and cause significant distress to providers and patients. It is, therefore, of great interest to healthcare organizations to predict what patients are at risk to be readmitted to their hospitals. However, current logistic regression based risk prediction models have limited prediction power when applied to hospital administrative data. Meanwhile, although decision trees and random forests have been applied, they tend to be too complex to understand among the hospital practitioners. OBJECTIVES: Explore the use of conditional logistic regression to increase the prediction accuracy. METHODS: We analyzed an HCUP statewide inpatient discharge record dataset, which includes patient demographics, clinical and care utilization data from California. We extracted records of heart failure Medicare beneficiaries who had inpatient experience during an 11-month period. We corrected the data imbalance issue with under-sampling. In our study, we first applied standard logistic regression and decision tree to obtain influential variables and derive practically meaning decision rules. We then stratified the original data set accordingly and applied logistic regression on each data stratum. We further explored the effect of interacting variables in the logistic regression modeling. We conducted cross validation to assess the overall prediction performance of conditional logistic regression (CLR) and compared it with standard classification models. RESULTS: The developed CLR models outperformed several standard classification models (e.g., straightforward logistic regression, stepwise logistic regression, random forest, support vector machine). For example, the best CLR model improved the classification accuracy by nearly 20% over the straightforward logistic regression model. Furthermore, the developed CLR models tend to achieve better sensitivity of more than 10% over the standard classification models, which can be translated to correct labeling of additional 400 - 500 readmissions for heart failure patients in the state of California over a year. Lastly, several key predictor identified from the HCUP data include the disposition location from discharge, the number of chronic conditions, and the number of acute procedures. CONCLUSIONS: It would be beneficial to apply simple decision rules obtained from the decision tree in an ad-hoc manner to guide the cohort stratification. It could be potentially beneficial to explore the effect of pairwise interactions between influential predictors when building the logistic regression models for different data strata. Judicious use of the ad-hoc CLR models developed offers insights into future development of prediction models for hospital readmissions, which can lead to better intuition in identifying high-risk patients and developing effective post-discharge care strategies. Lastly, this paper is expected to raise the awareness of collecting data on additional markers and developing necessary database infrastructure for larger-scale exploratory studies on readmission risk prediction.

Improving Hospital-Wide Early Resource Allocation through Machine Learning.

The objective of this paper is to evaluate the extent to which early determination of diagnosis-related groups (DRGs) can be used for better allocation of scarce hospital resources. When elective patients seek admission, the true DRG, currently determined only at discharge, is unknown. We approach the problem of early DRG determination in three stages: (1) test how much a Naïve Bayes classifier can improve classification accuracy as compared to a hospital's current approach; (2) develop a statistical program that makes admission and scheduling decisions based on the patients' clincial pathways and scarce hospital resources; and (3) feed the DRG as classified by the Naïve Bayes classifier and the hospitals' baseline approach into the model (which we evaluate in simulation). Our results reveal that the DRG grouper performs poorly in classifying the DRG correctly before admission while the Naïve Bayes approach substantially improves the classification task. The results from the connection of the classification method with the mathematical program also reveal that resource allocation decisions can be more effective and efficient with the hybrid approach.

Automated Classification of Clinical Incident Types.

We consider the task of automatic classification of clinical incident reports using machine learning methods. Our data consists of 5448 clinical incident reports collected from the Incident Information Management System used by 7 hospitals in the state of New South Wales in Australia. We evaluate the performance of four classification algorithms: decision tree, naïve Bayes, multinomial naïve Bayes and support vector machine. We initially consider 13 classes (incident types) that were then reduced to 12, and show that it is possible to build accurate classifiers. The most accurate classifier was the multinomial naïve Bayes achieving accuracy of 80.44% and AUC of 0.91. We also investigate the effect of class labelling by an ordinary clinician and an expert, and show that when the data is labelled by an expert the classification performance of all classifiers improves. We found that again the best classifier was multinomial naïve Bayes achieving accuracy of 81.32% and AUC of 0.97. Our results show that some classes in the Incident Information Management System such as Primary Care are not distinct and their removal can improve performance; some other classes such as Aggression Victim are easier to classify than others such as Behavior and Human Performance. In summary, we show that the classification performance can be improved by expert class labelling of the training data, removing classes that are not well defined and selecting appropriate machine learning classifiers.

Comparison of procedure coding systems for level 1 and 2 hospitals in South Africa.

The ability of three procedure coding systems to reflect the procedure concepts extracted from patient records from six hospitals was compared, in order to inform decision making about a procedure coding standard for South Africa. A convenience sample of 126 procedure concepts was extracted from patient records at three level 1 hospitals and three level 2 hospitals. Each procedure concept was coded using ICPC-2, ICD-9-CM, and CCSA-2001. The extent to which each code assigned actually reflected the procedure concept was evaluated (between 'no match' and 'complete match'). For the study sample, CCSA-2001 was found to reflect the procedure concepts most completely, followed by ICD-9-CM and then ICPC-2. In practice, decision making about procedure coding standards would depend on multiple factors in addition to coding accuracy.

Prospective, randomized evaluation of a personal digital assistant-based research tool in the emergency department.

BACKGROUND: Personal digital assistants (PDA) offer putative advantages over paper for collecting research data. However, there are no data prospectively comparing PDA and paper in the emergency department. The aim of this study was to prospectively compare the performance of PDA and paper enrollment instruments with respect to time required and errors generated. METHODS: We randomized consecutive patients enrolled in an ongoing prospective study to having their data recorded either on a PDA or a paper data collection instrument. For each method, we recorded the total time required for enrollment, and the time required for manual transcription (paper) onto a computer database. We compared data error rates by examining missing data, nonsensical data, and errors made during the transcription of paper forms. Statistical comparisons were performed by Kruskal-Wallis and Poisson regression analyses for time and errors, respectively. RESULTS: We enrolled 68 patients (37 PDA, 31 paper). Two of 31 paper forms were not available for analysis. Total data gathering times, inclusive of transcription, were significantly less for PDA (6:13 min per patient) compared to paper (9:12 min per patient; p < 0.001). There were a total of 0.9 missing and nonsense errors per paper form compared to 0.2 errors per PDA form (p < 0.001). An additional 0.7 errors per paper form were generated during transcription. In total, there were 1.6 errors per paper form and 0.2 errors per PDA form (p < 0.001). CONCLUSION: Using a PDA-based data collection instrument for clinical research reduces the time required for data gathering and significantly improves data integrity.

Managing health information during disasters: a survey of Victorian hospitals' current specialised health information systems.

It can be predicted that a substantial number of patients will seek medical care during a possible disaster, placing an increased strain on hospital resources, including health information services. With medical records playing a vital role in the identification of patients and documentation of patient care, the ability of the health information system to cope with this projected surge in demand needs to be addressed. This study was designed to investigate the expected use of specialised health information systems for disasters in Victorian hospitals during such contingencies. Specifically, this study investigated what type of specialised systems hospitals had in place at the time and whether a standard for specialised health information systems for disasters was needed. While 79% of responding hospitals reported having a specialised health information system for disasters, 91% of all responding hospitals reported that specialised health information systems for disasters were necessary. All specialised systems were paper-based, and 94% were based on the standard medical record format and content. Finally, 64% of hospitals believed that a Standard for specialised disaster medical records should be developed.

Taxonomy of health networks and systems: a reassessment.

OBJECTIVE: To assess a widely recognized multihospital system taxonomy. DATA SOURCES: The original taxonomy was based on American Hospital Association (AHA) Annual Survey Data for the years 1994 and 1995 and a reexamined version, on 1998 AHA data. STUDY DESIGN: We assess the appropriateness of using data designed to capture local hospital/system interrelationships to develop a taxonomy of multihospital systems. DATA ABSTRACTION METHODS: The original and reexamined taxonomies used dichotomous measures of service availability, physician practice ownership, and managed care offerings. PRINCIPAL FINDINGS: The data and measures used to formulate the taxonomy are not appropriate for classifying multihospital systems at the company level. CONCLUSIONS: Taxonomic studies of multihospital systems are very much needed; future taxonomic studies should make clear distinctions between systems at local versus company levels.

Current and future use of ICT for patient care and management in German acute hospitals--a comparison of the nursing and the hospital managers' perspectives.

OBJECTIVES: This study aimed at gaining comprehensive information on the current status of patient care and management applications used in German acute hospitals. Since the degree of ICT coverage in hospitals depends on the attitude of the key decision makers we also wanted to capture their plans and priorities and herewith try to predict future use. METHODS: We therefore conducted a nation-wide survey including all acute hospitals in Germany in which two questionnaires were mailed to each hospital, one to the nursing managers, the other to the hospital managers. RESULTS: Six hundred hospitals participated in the survey which corresponds to an overall response rate of 27.6%. Accounting (84%) was found to be the most prevalent management module. Rostering was implemented in every second hospital. For clinical applications laboratory systems ranked first (69%). Ordering systems were used in nearly every second hospital. Nineteen percent of the hospitals reported employing an electronic patient record, 7% a nursing documentation system. Ranked by their priorities ordering systems hold the first position and care planning the last position. According to their plans, hospital managers, not nursing managers, intend to introduce nursing documentation. In contrast, nursing managers favor ordering and rostering for the near future. CONCLUSIONS: There is still a preponderance of management-oriented systems in German hospitals, yet clinical applications, in particular those supporting communications, will gain ground. The future of documentation systems is unclear, unless they not only provide statistical data for the management but support the clinical process properly.

Health care provider use of private sector internal error-reporting systems.

The purpose of this study was to review the state of the art of private sector internal error-reporting systems and to begin to develop a classification system for comparing systems. Interviews were conducted to research and examine 9 systems currently on the market. Analysis resulted in the following observations: (1) 7 of the systems are stand-alone, while 2 are part of larger hospital information systems; (2) most of the systems have been in existence for less than 5 years; (3) acute care hospitals are the primary clients; (4) systems are capable of interfacing with other information systems and root-cause analysis programs; and (5) systems are browser based and accessible via the Internet and/or the provider's intranet. Additional studies are needed to determine the impact of these systems on health outcomes. However, one fact is clear: tracking incidents will not improve patient safety unless administrators close the feedback loop on quality.

The aging of a clinical information system.

The senescence of a clinical information system is more likely to have administrative than technical bases. Supporting this claim is a case study of one aging oncology information system. The case study is qualitative, as behooves the subject matter. Content analysis of several documents suggests that the change in job description of the data coordinator led to a workflow breakdown. Next, twenty-two individuals were interviewed. Notes from the interviews were coded, and the resulting patterns led to partial support for the workflow breakdown conjecture, refutation of the hypothesis that users disliked the character-based, human-computer interface, support of the conjecture that political rather than technical factors drive the usage patterns of the system, and evidence that 'political' activity will determine the future of the information system. A stakeholder matrix is proposed that addresses administrative concerns. Also, the issue of the uniqueness of any oncology clinical information system is linked to the plans for this legacy system.

Superiority is complex.

Does being Most Wired make a difference? It's a big question to answer. The evidence suggests that the Most Wired hospitals outperform their peers. But do the numbers tell the whole story?

The 100 most wired 2004.

Even as they speed ahead in their quest to improve quality and operations, the top IT hospitals are shifting gears away from a fascination with whiz-bang technology to the nitty-gritty work that ensures success. In the foldout, you'll find the 100 Most Wired, Innovator Award winners, Most Wired-Small and Rural, Most Wireless and Most Improved.

Most wired 2003.

A record number of health care organizations are represented in H&HN's fifth Most Wired Survey and Benchmarking Study. This year's survey reveals numerous trends in IT strategies among the 100 Most Wired, and the winners of the Innovator, Most Improved and Most Wireless awards. New this year: the Most Wired Small and Rural Award.

Information technology. For IT staffers, one happy workplace.

Hospitals are saddled with the reputation of not attracting the best and brightest information technology talent. It may be a bad rap; one Kansas City hospital stands out as one of the best places to work in IT in the country.