Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
A practical data mining method to link hospital microbiology and an infection control database Thanh Kim Dao, MSHS, CIC, Firas Zabaneh, MT, CIC, Judy Holmes, MT, CIC, Laura Disrude, MT, Margaret Price, PhD, and Layne Gentry, MD Houston, Texas Background: Data mining is the process of data exploration, selection, and transformation. Data mining has developed as an important tool in health care for converting immense amounts of data into powerful information to support surveillance, detect and assist with investigation of microbial clusters, and collaborate data for research. Additionally, data mining supports ICPs’ commitment to providing quality health care. Methods: Realizing the existing potential of a current infection control database, current software has been transformed into a system capable of data mining with the addition of a single laboratory information system interface. Results: Commercial data mining software products are available but at significant cost. With a laboratory interface to a current infection control database, a system capable of data mining was created. Similar to other data mining products, this system achieves electronic surveillance, generates monthly or more frequent reports as required, provides patient diagnosis and culture history, offers unit-specific rates, and provides rapid identification of patients with multidrug-resistant organisms. Additionally, health care system-wide custom report-writing features are highly effective tools used to assist with outbreak investigations and research projects. Conclusion: Purchasing new software products can be expensive; however, current resources may be used to meet data mining needs. Although weekly queries are optimal, data mining software must provide, at a minimum, monthly reports to track infection-related events in a health care facility. The capability to detect a sudden increase in any specific microorganism and the ability to monitor a known problematic microorganism is a necessity. In addition, data mining software products must include graphing capabilities and ease of use custom report-writing features. (Am J Infect Control 2008;36:S18-20.) Data mining is a process of data exploration, selection, and transformation. This process can be concurrently or retrospectively applied to immense amounts of data stored in databases in an automated fashion that incorporates pattern recognition and data visualization to obtain knowledge.1 Using analytical techniques and statistical methods to identify significant patterns and trends, data mining and statistics together produce synergy.2 Data mining surveillance technology explores microbiology data across the interface to select for significant data and transform data into useful information for ICPs. Data mining is considered an adjunct to infection control quality and does not replace routine surveillance; ICPs have the obligation to conduct routine surveillance based on Centers for Disease Control and From the Department of Infection Control, Saint Luke’s Episcopal Hospital, Houston, TX. Address correspondence to Thanh Kim Dao, MT, CIC, Department of Infection Control, Saint Luke’s Episcopal Hospital, 6720 Bertner Avenue, M/C 1-166, Houston, TX 77030. E-mail: [email protected]. 0196-6553/$34.00 Copyright ª 2008 by the Association for Professionals in Infection Control and Epidemiology, Inc. doi:10.1016/j.ajic.2007.05.010 Prevention (CDC) definitions of health care–associated infections.3 Data mining has developed as an important tool in health care for converting enormous amounts of data into powerful information to enhance surveillance, detect and assist with investigation of microbial clusters, and support research.4,5 Because infection control surveillance is based on microbiology culture results and patient admission information, the addition of a single interface from the laboratory information system (LIS) to an existing infection control database has the potential to significantly impact ICPs daily workload by saving time and improve overall quality. Whereas the cost of buying commercial data mining software can be significant, the addition of a single LIS interface to an existing infection control database provides similar capabilities for a minimal fee. Standard reports from this automatic LIS data feed automate disease reporting, provide automated alerts for pertinent microbiological events and early warnings, and allow for onsite and offsite surveillance capabilities. METHODS Infection control database requirements For effective transformation of an existing infection control software system into one capable of data S18 FLA 5.0 DTD ymic1203 11 March 2008 1:33 am ce 23 Dao et al April 2008 mining, the current infection control database must include the following: automatic rate calculations; standard report formats; ease-of-use custom reportwriting capabilities; graphic functions; and an admit, discharge, and transfer (ADT) interface to receive patient admission information in real time. The addition of an LIS interface to receive preliminary and final microbiology data in real time is preferable. At minimum, microbiology information should come across the interface at least once a day (daily batch.) Microbiology data requirements Microbiology data received should be encounterbased and include the following: ordering physician, specimen source, culture collection date and time, culture type, culture results, organism identification, and collection location. Susceptibility patterns may be included or omitted if inclusion hinders or limits project progression. Susceptibility is considered an added benefit for advance data mining to further assist with outbreak investigation but not a necessity for basic data mining purposes. Effective data transmission If the sending system is compatible with free text, data should be received as text to ensure all transmitted data are received; however, terminology from a predefined dictionary in the receiving system is frequently used for ease of report-writing for the nonclinical system administrator. In this case, the sending system administrator must successfully communicate all dictionary changes to the receiving system administrator. For example, if the LIS adds ‘‘ventricular assist device’’ to its culture source dictionary, but the receiving system’s culture source dictionary is not updated, any cultures with ‘‘ventricular assist device’’ as a culture source will not populate or will show as ‘‘null’’ because that terminology was not recognized by the receiving system. Traditionally, microbiology departments prefer to document culture source as precisely as possible; therefore, standard dictionary terms may be used in conjunction with free text. In most cases, culture source should be received as text. Implementation team member requirements The project team must include a microbiology data expert, an infection control practitioner, LIS and infection control system administrators, and interface expert team members. The microbiology team member must be an internal candidate who is familiar with the microbiology department documentation processes. Project development may also benefit from having an external informatics consultant who serves as a clinical–technical liaison. S19 Allocation of enough resources and time to thoroughly build and test the interface before implementation is a necessity to ensure data accuracy and is critical to project success. The receiving system administrator must take an active role in validating interface transactions and periodically monitor for data integrity crossing the interface. Microbiology culture data are unique and complex. Unlike other laboratory test results, the combination of different microorganisms, range of antibiotics and susceptibilities create a multitude of possible combinations that continue to be a challenge for database designers and for information technology as a whole. RESULTS Capabilities of this infection control data base/LIS data mining system include automatic rate calculation and early detection of an increased rate through its graphical functionalities. For example, this system automatically calculates and graphs positive wound, blood, or urine cultures per 10,000 patient days or inpatient admissions. These infection rates analyzed with a statistical software packet such as QI Macro can facilitate ‘‘early warnings’’ of a process out of control, identify significant variation in pattern of organisms, or signal a ‘‘pertinent’’ microbiological event warranting further investigation. Data mining is not just data processing. It is a systematic method to identify excessive rates or unusual patterns of positive microbiology cultures based on patient location, culture source, and organism identification.6 The availability of culture results in electronic format allows for onsite and offsite CDC definition– based surveillance. Furthermore, the capability of a data mining system to export data to Excel can streamline surveillance and increase turnaround time for ICPs. Additionally, this system has the potential to produce unit-specific multidrug-resistant organism rates and health care–associated infection rates, used by nursing and administration as a quality indicator marker. Reportable conditions reporting to local and state health departments can also be automated. Information from LIS in conjunction with patient information from ADT is combined to generate this report. From the ADT interface, patient demographics are merged with infection-related data to provide records in an electronic format for infection prevention research purposes. In addition, the ability of the system to graph data permits effortless visual interpretation of data. DISCUSSION Data mining is a powerful tool for ICPs, but data output is equivalent to data input; therefore, data quality is FLA 5.0 DTD ymic1203 11 March 2008 1:33 am ce 23 S20 Dao et al Vol. 36 No. 3 Supplement 1 a key requirement for data mining.1 Monitoring interface transactions is a standard component of any interface-related project. This identifies the number of records rejected requiring further follow-up and is a measure of completeness of data transfer. Additional validation of data crossing the interface should be conducted periodically to ensure accuracy and comprehensiveness of the data transferred. for busy ICPs. Early recognition is the key to prevention of outbreaks, and weekly trending reports allow this. Effective data mining permits microbial pattern recognition and detects the presence of microbial clusters that warrants early investigation to enhance patient safety and prevent costly outbreaks. References CONCLUSION A data mining system created by the addition of a single LIS interface to a standard infection control database has successfully been designed and implemented at minimal cost to the hospital. This system meets data mining needs of a health care system infection control department that includes an 888-bed hospital and its 90-bed community hospital. Data visualization with graphics is the most useful form of data exploration. This data mining system is capable of providing trending reports and detecting sudden increases in specific pathogens using established queries. Ease of use is the key for data mining success, and this system is extremely user-friendly 1. Stilou S, Bamidis P, Maglaveras N, Pappa S. Mining association rules from clinical databases: an intelligent diagnostic process in healthcare. Medinfo 2001;10:1399-403. 2. Obenshain M. Application of data mining techniques to healthcare data. Infect Control Hosp Epidemiol 2004;25:690-5. 3. Scheckler W, Brimhall D, Buck AS, Farr BM, Friedman C, Garibaldi RA, et al. Requirements for infrastructure and essential activities of infection control and epidemiology in hospitals: a consensus panel report. Available from: http://www.apic.org/AM/Template.cfm?Section5Search §ion5Consensus_Reports&template5/CM/ContentDisplay.cfm& ContentFileID5284. Accessed March 1, 2007. 4. Koh HC, Tan G. Data mining application in healthcare. J Healthc Inf Manag 2005;19:64-72. 5. Berger AM, Berger CR. Data mining as a tool for research and knowledge development in nursing. Comput Inform Nurs 2004;22:123-31. 6. Schifman RB, Palmer RA. Surveillance of nosocomial infections by computer analysis of positive culture rates. J Clin Microbiol 1985;21:493-5. FLA 5.0 DTD ymic1203 11 March 2008 1:33 am ce 23