Development of an anesthesia data warehouse: Preliminary results

Development of an anesthesia data warehouse: Preliminary results

Disponible en ligne sur ScienceDirect www.sciencedirect.com IRBM 34 (2013) 376–378 Technical note Development of an anesthesia data warehouse: Prel...

479KB Sizes 0 Downloads 44 Views

Disponible en ligne sur

ScienceDirect www.sciencedirect.com IRBM 34 (2013) 376–378

Technical note

Development of an anesthesia data warehouse: Preliminary results A. Lamer ∗ , M. Jeanne , B. Vallet , G. Ditilyeu , F. Delaby , B. Tavernier , R. Logier Inserm CIC-IT 807, University Hospital, Institut Hippocrate, 2, avenue Oscar-Lambret, 59037 Lille, France Received 16 May 2013; received in revised form 16 September 2013; accepted 18 September 2013 Available online 18 October 2013

Abstract Hospital information system manages patient’s hospitalization information across different applications and databases. As statistics are performed with difficulties on these different databases, the university hospital of Lille developed an anesthesia data warehouse. This common structure stores data related to anesthesia procedures and patient hospital stay. In that way, the joint analysis on intervention’s events and patient’s outcome is possible. However, data quality remains one of the main issues in this kind of project. Indeed, errors in patient identifier result in difficulties to link data between the different sources. This problem will be approached in the next phase of the project. © 2013 Elsevier Masson SAS. All rights reserved.

1. Introduction The hospital information system (HIS) records and stores information related to patient’s hospitalization. The HIS is composed of a set of software and allows to each medical unit across the hospital site to access to patient data. In particular, HIS contains data about patient’s hospital stay: • hospital stay duration; • scheduled or emergent hospitalization; • type of care unit: intensive care unit, continuous care or standard care unit; • whether death occurs during hospital stay; • all medical procedures and diagnostics; • lab results. The anesthesia information management system (AIMS) [1] is part of the HIS and manages information about all anesthesia procedures. It is composed of different modules, which collect and centralize all data referring to one case, from pre-anesthetic evaluation to discharge from post-anesthesia care unit (PACU). For example, patient identity, preoperative comorbidities, clinical parameters as heart rate, blood pressure or respiratory rate, administered drugs and main events during anesthesia and surgery are recorded by the AIMS. ∗

Corresponding author. E-mail address: [email protected] (A. Lamer).

1959-0318/$ – see front matter © 2013 Elsevier Masson SAS. All rights reserved. http://dx.doi.org/10.1016/j.irbm.2013.09.005

Even if AIMS is integrated in the HIS, it is difficult to perform statistics on data from AIMS and from other databases. However, each database of the HIS has its own structure and links between them are not fully explicit. Moreover, such databases have a transactional structure that is well adapted for a daily use but not really suited for statistical analysis. Currently, postoperative mortality (non-cardiac, nonneurosurgical, non-ambulatory and non-obstetrical) remains high in Europe and in France [2]. As we previously contributed to demonstrate in the European EuSOS study (European Surgical Outcome Study) [2], mortality remains around 3% in France, below the European rate of 4%. A statistical association exists between non-compliance with quality-of-care indicators and patient outcome. Recent literature in the field [3,4] suggests that perioperative patient care improvement based on a rigorous enforcement of compliance with quality-of-care indicators, may contribute to decrease mortality to 1% or perhaps less than 0.5%. In this way, AIMS may be used to detect per operative adverse events as hypotension, tachycardia, low entropy, etc. and HIS has information about patient outcome, mortality and morbidities as abnormal length of stay, unplanned admission to critical care, myocardial infarction, cardiac arrest, hemorrhage. . . It would be useful to identify links between per operative adverse events (from AIMS) and patient outcome (from HIS). In our institution, the HIS and AIMS have the necessary information to perform this kind of study. That is why, in order to perform joint analysis on AIMS and the rest of the HIS, we developed an anesthesia data warehouse [5,6]. A data warehouse is a common structure in which data from different

A. Lamer et al. / IRBM 34 (2013) 376–378

377

Fig. 1. Feeding process of a data warehouse.

Fig. 2. Data model of the data warehouse.

sources are centralized. The data warehouse is fed via the ETL (Extract, Transform, Load) process: data are first extracted from the source systems, then data are transformed, cleaned and aggregated in a way to facilitate analysis, finally data are loaded in the data warehouse (Fig. 1). As a result, information initially registered and stored in separate systems for different purposes, is brought together in a common repository and may be jointly analyzed.

Business Object XI 3.1SP5. Statistics are conducted with free software environment R. One of the main objectives of the ETL process is to link patients and interventions from AIMS with hospital stays from HIS. For this, the unique patient identifier, stored in the two systems, and the dates of intervention and hospital stays are used.

2. Materials and methods

3. Results

The software CORA-PMSI [7], developed by the editor McKesson (San Francisco, United States) is used since 2010 in the university hospital of Lille and provides detailed information about 150,000 patient’s hospital stay each year. The AIMS DIANE® [8] edited by Bow Medical (Amiens, France) is used since 2005 and records data from about 55,000 anesthesia procedures a year. This corresponds to around 200 millions vital signs and 4 millions drug administration or intervention steps recorded by the AIMS each year. As the software CORA-PMSI is used since 2010, only interventions of the last three years may be linked with hospital stay information. We potentially may analyze about 160,000 interventions corresponding to 110,000 patients. The anesthesia data warehouse is implemented with information from these two data source systems. A data model has been established. The Fig. 2 represents a simplified version of this data model. The tables “hospital stay” and “detailed stay” are fed with information from CORA-PMSI while the tables “patient”, “intervention”, “measurements” and “events” are stored data from Diane® . The data warehouse is stored on an oracle database in the version 11.2. The ETL process was conducted with OpenText Integration Center (Genio 7.1), edited by OpenText. Some specific applications for data cleaning are developed using Microsoft. NET environment. Reports are edited with SAP

The ETL process is now launched routinely each week and collects data from the last two weeks. The weekly treatment goes on for around 7 hours, and load data from about 2000 interventions. Over the 7 hours, 5 hours are due to measurements loading. The high volume of data is one of the critical point of the ETL process. A special process had to be developed and launched for recovery period of one month over the past three years. We now dispose of information about more than 150,000 interventions in total. Fig. 3 represents the number of interventions per month, for each year, which are now available in the data warehouse. Due to imprecision in manually entering of patient identity in the different systems, around 5000 interventions failed to be linked with the hospital stay. Indeed, links between data from HIS and from AIMS are established with the unique patient identifier. However, some patients in AIMS are missing this information. For the linked interventions, statistics may be calculated. As an example, the median hospital stay length was calculated per ASA status [9]. The results are available in Fig. 4. This kind of result is representative of the interest of the data warehouse. Indeed this query needs information from the two data sources. The patient’s ASA status for the intervention is collected by the AIMS while the hospital stay length is stored by others applications in the HIS.

Fig. 3. Number of interventions per month over the last three years.

378

A. Lamer et al. / IRBM 34 (2013) 376–378

Fig. 4. Example of analysis performed on the anesthesia data warehouse, hospital stay length in function of the ASA status.

4. Discussion – conclusion A data warehouse allows performing statistics on data from different source systems. In the case of the university hospital of Lille, we developed a data warehouse, which stores information related to anesthesia procedures and hospital stay. This allows conducting joint analysis on the data from two different sources: an AIMS and hospital stay management system. The data warehouse is actually used for a medico-economic purpose, but also for clinical research. Some studies are performed to assess the quality of anesthesia and its impact on patient outcome. Two main difficulties are actually encountered. The first is linked to the high volume of data. As the number of measurements will increase week after week, optimization operations will have to be performed, in order to keep fast response time when loading and querying the data warehouse. During the ETL processes, SQL procedures may be used to reduce loading time. Indeed this kind of procedure will only work on the database engine, instead of usual ETL processes, which will extract data from the database, and apply transformations on the ETL engine, before loading the database. During the querying operations, data structures as indexes and partitions may be used in order to keep fast response. The second one is related to data quality as links between each intervention and hospital stay have not been established for all anesthesia procedures, due to incomplete patient identification in AIMS. Each patient normally has identity information as a unique identifier common to all the databases in the institution. When this identifier is missing, other information as name,

forename, birth date, birth place can be used to look for the right patient in the others sources. Data cleaning and data linkage [10] are regular operations in data warehousing. It’s the subject of current activity, in order to enhance the quality of the data warehouse. References [1] Muravchick S, Caldwell JE, Epstein RH, Galati M, Levy WJ, O’Reilly M, et al. Anesthesia information management system implementation: a practical guide. Anesth Analg 2008;107:1598–608. [2] Pearse RM, Moreno RP, Bauer P, Pelosi P, Metnitz P, Spies C, et al. Mortality after surgery in Europe: a 7-day cohort study. Lancet 2012;380(9847):1059–65. [3] Sessler DI, Sigl JC, Kelley SD, Chamoun NG, Manberg PJ, Saager L, et al. Hospital stay and mortality are increased in patients having a “triple low” of low blood pressure, low bispectral index, and low minimum alveolar concentration of volatile anesthesia. Anesthesiology 2012;116(6):1195–203. [4] Kertai MD, Palanca BJ, Pal N, Burnside BA, Zhang L, Sadiq F, et al. Bispectral index monitoring, duration of bispectral index below 45, patient risk factors, and intermediate-term mortality after noncardiac surgery in the B-Unaware Trial. Anesthesiology 2011;114(3):545–56. [5] Kimball R, Reeves L, Ross M, Thornthwaite W. Le data warehouse, guide de conduite de projet. Paris, France: Eyrolles; 2005. [6] Li P, Wu T, Chen M, Zhou B, Xu W. A study on building data warehouse of hospital information system. Chin Med J 2011;124:2372–3277. [7] McKesson – France: https://www.mckesson.fr/. [Accessed 6th February 2012]. [8] Bow Medical: http://www.bowmedical.com/. [Accessed 6th February 2013]. [9] ASA status, wikipedia: http://en.wikipedia.org/wiki/ASA physical status classification system. [Accessed on 5th February 2013]. [10] Winkler W. Overview of record linkage and current research directions. U.S. Census Bureau, Technical Report; 2006.