a Dipartimento di Informatica e Sistemistica, Università di Pavia, via Ferrata 1, 27100 Pavia, Italy
b Unità di Nefrologia e Dialisi, Azienda Ospedaliera di Vigevano,Italy
* The corresponding author [email protected]
Abstract
The aim of this paper is to describe a research project that deals with the definition of tools for the quality assessment of hemodialysis centers, on the basis of the data automatically collected during dialysis sessions. The ultimate goal of this project is to provide health care professional with a system to be used in clinical practice; such system should be able to highlight the performance of the overall dialysis center in terms of patients’ outcomes, efficiency of the dialysis procedures and number of personnel interventions during hemodialysis sessions. Moreover, it should assist health care providers in discovering the causes of negative clinical results. A combination of simple descriptive techniques, rule extraction and temporal reasoning is used to accomplish with this goal. The paper briefly describes the application domain, the basic goals of the project and the software that has been now made available to the physician responsible of a limited assistance dialysis centers. Some clinically interesting results, obtained on data coming from 5 patients monitored for nine months, are also shown.
1. Introduction
The growing interest on knowledge management within health care institutions have highlighted the crucial role of data collected in the clinical routine for organizational learning [1]. In particular, a crucial aspect of organizational learning is represented by the
assessment of the quality of a hospital service [2-3].
In this paper we are interested in the definition of a software tool for assessing the efficacy of treatment delivered by a Hemodialysis Department (HD) on the basis of the data routinely collected during hemodialysis sessions. Rather interestingly, the most recent hemodialyzers allow for the automatic monitoring of the dialysis sessions, and are also equipped with network boards, so that it is possible to simultaneously collect data coming from all the concurrent dialysis sessions running in a HD center.
Although full monitoring of dialysis sessions has been conceived as a way to intervene during or immediately after each session, the use of those kind of data are often routinely neglected. Emergencies are managed by the nurses in charge of the dialysis session, and the personnel do not have time even to visually inspect each session. It is therefore necessary to rely on tools able to automatically synthesize the information contained in the monitoring data, allowing the health care providers to periodically evaluate the HD performance, either for what concerns all patients or for what concerns each patient. This goal can be reached by developing an auditing system, able to summarize the dialysis sessions from a clinical quality viewpoint [4].
2. Assessing hemodialysis outcomes
The main idea of the system we are working on is the
capability of assessing the most widely recognized dialysis outcomes indicators on the basis of the variables automatically collected by the hemodialyzers. In particular, we have considered: i) the efficiency of the removal of protein catabolism products (urea, creatinine), evaluated through the measurement of the blood flow in the extracorporeal circuit (QB), the body weight loss (WL) and the dialysis time (T); ii) the efficiency of the extra-corporeal circuit of the dialyzer, measured through the pressure of the circuit before (arterial pressure, AP) and after (venous pressure VP) the dialyzer; iii) the Body water reduction and hypotension episodes, performed measuring the body weight (BW), systolic and diastolic pressures, the cardiac frequency, the hemoglobin concentration and the Hematic Volume. Such measurements have been further summarized by extracting, for each dialysis session, their median, 10th and 90th percentile.
Relying on this extracted variables, we have defined the following outcome parameters: i) the median levels of QB, VP, AP, ii) the time difference between the prescribed dialysis time and the effective one ( time), iii) the difference between the prescribed weight loss and the real loss at the end of the dialysis ( W), iv) the difference between the ideal weight of that patient (called Net Weight) and the real weight obtained at the end of the dialysis ( L), iv) the presence of a nurse intervention during each dialysis session. A hemodialysis session may be considered as failed, when at least one outcome parameter is out of a predefined range, defined by the HD responsible.
3. The HEMOSTAT software
Following the ideas described in Section 2, we have implemented a software tool, called HEMOSTAT, written in Java 1.2 and based on MySQL data base;
HEMOSTAT works on the data collected by the DIALMASTER software program, provided by the HOSPAL company together with the INTEGRA dialyzers. Such devices are able to monitor up to 50 variables with sampling time typically equal to 1 min.
HEMOSTAT is a complete auditing tool, that allow to define the hemodialysis targets for each outcome parameters for each patient, and to easily calculate the percentage of failures for the overall HD and for each patient, over any given monitoring time period.
Moreover, it provides histograms, pies, charts and text outputs.
4. HEMOSTAT at work
The HEMOSTAT software is under testing in the limited assistance center located in Mede, Italy, and managed by the Unit of Nephrology and dialysis of the Hospital of Vigevano, Italy. Up to now, we have tested the methodology presented above on the data set coming from 5 patients monitored for about nine months, for a total of 383 dialysis sessions with a complete set of data. Each of the monitored variables was sampled every minute. Table 1 shows a synthesis of the dialysis center performance, in terms of success of the different outcomes considered. The results show that in the 47% of cases the outcomes were out of target. In the majority of the cases the failure was due to L (23% of sessions, 48% of failures), to the insufficient weight loss at the end of the session (21% of the sessions, 44% of the failures) and to the Bulk Blood Flow (18% of sessions, 37% of failures). Several times multiple failures occurs: the analysis of the co-occurrences shows that QB, W, L are associated with another one cause of failure in 28 cases over 69 (41% of cases), 42 cases over 82 (51% of cases) and in 37 cases over 89 (42% of cases), respectively. Moreover, they occurred 9 times together.
76
Table 1. Outcomes assessment
Parameters Failures Successes
VP 2 381
QB 69 314
AP 0 383
time 6 377
W 82 301
L 89 294
Overall 187 196
Our project is not only dealing with the comparison of median values with predefined goals, but it is also studying the automatic discovery of relationships between failures and monitoring variables. To this end, we are investigating the use of temporal abstractions [5] and confirmation rules [6] to derive associations between time patterns and failures, either between and within dialysis sessions. Through these instruments it has been possible, for example, to look for the main reasons of failures related to the L parameter. In this case, the following rule has been discovered: IF WB-trend is not steady and WB is High THEN L fail; where WB is the weight at the beginning of dialysis. Such rule finds the cause of L failure in the instability of the body weight control between dialysis sessions.
5. Future work
The HEMOSTAT software is undergoing an on-site clinical evaluation; a new release of the software, equipped with data mining techniques, is foreseen by July 2002. In the next phase of the study all the day hospital beds will be monitored, and the number of patients will increase from 5 to 30. This will allow to draw better conclusions on the HD performance and to improve the quality of the results obtained through the approach herein presented.
Acknowledgements
This work is part of the project “Analysis, Information
Visualization and Visual Query in Databases for Clinical
Monitoring”, funded by the Italian Ministry of Education.
References
[1] Abidi SS. Knowledge management in healthcare: towards
'knowledge-driven' decision-support services. Int J Med Inf.
2001 Sep;63(1-2):5-18.
[2] Zoccali C. Medical knowledge, quality of life and
accreditation of quality in health care. The perspective of the
clinical nephrologist. Int J Artif Organs. 1998
Nov;21(11):717-20.
[3] Eknoyan G, Levin NW, Steinberg E. The National Kidney
Foundation Dialysis Outcomes Quality Initiative. Curr Opin
Nephrol Hypertens. 1997 Nov;6(6):520-3.
[4] Bellazzi R, Magni P, Bellazzi R. Improving dialysis services
through information technology: from telemedicine to data
mining. Medinfo. 2001;10(Pt 1):795-9.
[5] Bellazzi R., Larizza C., Riva A., Temporal Abstractions for
Interpreting Diabetic patients monitoring data, Intelligent Data
Analysis, 2 (1998) 97-122.
[6] Gamberger D. and Lavrac N. Confirmation rule sets. In
Proc. of 4th European Conference on Principles of Data
Mining and Knowledge Discovery (PKDD 2000), 2000,
Springer, pp. 34-43.