A prognostic model for temporal courses that combines temporal abstraction and case-based reasoning

https://doi.org/10.1016/j.ijmedinf.2004.03.007Get rights and content

Summary

Since clinical management of patients and clinical research are essentially time-oriented endeavours, reasoning about time has become a hot topic in medical informatics. Here we present a method for prognosis of temporal courses, which combines temporal abstractions with case-based reasoning. It is useful for application domains where neither well-known standards, nor known periodicity, nor a complete domain theory exist. We have used our method in two prognostic applications. The first one deals with prognosis of the kidney function for intensive care patients. The idea is to elicit impairments on time, especially to warn against threatening kidney failures. Our second application deals with a completely different domain, namely geographical medicine. Its intention is to compute early warnings against approaching infectious diseases, which are characterised by irregular cyclic occurrences. So far, we have applied our program on influenza and bronchitis. In this paper, we focus on influenza forecast and show first experimental results.

Introduction

Since clinical management of patients and clinical research are essentially time-oriented endeavours, reasoning about time has become a hot topic in medical informatics. Here we present a method for prognosis of temporal courses based on numeric values, which we have applied on multiparametric kidney functions and on incidences of the spread of infectious diseases.

Since traditional time series techniques [1] work well with known periodicity, but do not fit in domains characterised by possibilities of abrupt changes, much research has been performed in the field of medical temporal course analysis in the recent years. However, usually either extensive domain theories or well-known standards (e.g. course pattern or periodicity) are required.

An ability of RÉSUMÉ [2] is the abstraction of many parameters into one single parameter and to analyse courses of this abstracted parameter. However, interpretation of these courses requires extensive domain knowledge. Haimowitz and Kohane [3] compare many parameters of current courses with well-known standards (trend templates). In VIE-VENT [4] both ideas are combined: courses of single quantitative measured parameters are abstracted into qualitative course descriptions that are matched with well-known standards.

When we started building a system for course analysis and prediction of the kidney function, we were confronted with a domain where the domain theory is extremely incomplete and no standards are known yet. So we had to design our own method. Because of our good experience with case-based reasoning (CBR) methods for medical applications [5] (especially for antibiotics therapy advice [6] and for the diagnosis of dismorphic syndromes [7]), we decided to use case-based reasoning again. For temporal courses, our basic idea is to search for former similar courses and to consider their course continuations as prognosis.

In this paper, at first we briefly explain the methods of temporal abstraction and of case-based reasoning. Subsequently, we introduce our prognostic model and present both applications, on kidney functions and on influenza.

Section snippets

Methods

Since our prognostic method combines temporal abstraction with case-based reasoning, here we briefly introduce both methods.

Our prognostic method

The case-based reasoning cycle developed by Aamodt and Plaza consists of retrieving former similar cases, adapting their solutions to a current problem, revising a proposed solution, and retaining new learned cases [9]. Fig. 1 shows an adaptation of this cycle to our medical temporal prognostic method.

Since in both applications, the idea is to give information about a specific development and its probable continuations, we do not generate a solution that should be revised by a user. So, in

Prognosis of kidney function courses

After presenting our prognostic model we want to show its use. The first application deals with the prognosis of kidney function courses of intensive care patients.

When we started the project physicians at our ICU got daily printed renal reports from the monitoring system NIMON [14]. Each report consisted of 13 measured and 33 calculated parameters of those patients where renal function monitoring was applied. The interpretation of all reported parameters was quite complex and special knowledge

Influenza forecast

The general goal of our TeCoMed project is to discover regional health risks in the German federal state Mecklenburg, Western Pomerania. So, we have developed a program that computes early warnings against forthcoming waves or even epidemics of infectious diseases that are characterised by cyclic but irregular occurrences. So far, we have applied our method to forecast influenza and bronchitis. Here we focus on influenza, for the application on bronchitis, see [20].

Many people believe influenza

Conclusions

In this paper, we have proposed a prognostic method for temporal courses, which combines temporal abstractions with case-based reasoning. We have applied our method on the prognosis of kidney function courses and on the prognosis of the spread of infectious diseases, especially of influenza. Though there are some differences between both applications, the main principles are the same. Temporal courses can be characterised by domain-dependent trend descriptions. The parameters of these

Acknowledgements

The research on kidney function prognoses was funded by the German Ministry for Research and Technology (BMBF). The research on influenza forecast is partly funded by the German Research Society (DFG).

References (28)

  • A. Aamodt et al.

    Case-based reasoning: foundation issues. Methodological variation and system approaches

    AI Commun.

    (1994)
  • J. Kolodner

    Case-based Reasoning

    (1993)
  • R. Macura, K. Macura, MacRad: radiology image resources with a case-based retrieval system, in: Proceedings of the...
  • R. Schmidt et al.

    Multiparametric time course prognoses by means of case-based reasoning and abstraction of data and time

    Med. Inform.

    (1997)
  • Cited by (0)

    View full text