Drug therapy decision support for infectious diseases showcased as prototypical Grid killer app

Lyon 16 January 2003At the University of Amsterdam, a distributed medical support system for drug therapy of HIV infection has been developed and is currently tried out in 25 hospitals and in 20 physician practices. Peter Sloot presented the rule-based expert system which was recently enhanced with a simulation unit, at the HealthGrid conference. Physicians can have fast access to the system through their PDA, WAP or mobile phone devices and via wireless networks. There are two ways to retrieve results quickly, whether by using the middleware approach, or by applying Java technology.


There are some fifteen accepted drugs for HIV but no doctor exactly knows which one to prescribe for a particular patient at a given moment. The team of Dr. Sloot addresses the problem from a multi-angled view where genomics, proteomics and immunology are taken into account at the three levels of the DNA, the protein and the cell structure in an attempt to move from molecule to man and discover the right drug for the right treatment. The huge knowledge of data involved has taken years to mature in scientific journals.

The idea is to benefit from this enormous amount of information, to take it and check it against what is known about a certain drug in order to insert it into a database, as the speaker explained. The mechanism of HIV cell invasion occurs as follows. The HIV invades and destroys the CD4 T lymphocytes by releasing its contents, consisting of 3 important enzymes and two strands of RNA. The reverse transcriptase (RT) copies the HIV RNA into double-strand DNA. The integrase (I) inserts the HIV DNA into a chromosome in the host cell and the protease (P) cuts the new protein molecules into a functional form.

A medical decision support system advising which drugs can be applied in an individual patient case of HIV infection has to contain several features. One of them is mesoscopic simulation using high performance computing to predict the virus behaviour for the near future. Another one is parameter space exploration for which high throughput computing is needed. Third, database federation and integration are needed for the data disclosure and parameter transfer for the data fusion. For visualising and accessing the data, a mobile device or PDA is enhanced with roaming and remote protocols.

Dr. Sloot showed how the emerging Grid technology adds real value to the system because it allows to seamlessly integrate the data on request, process the data mining with compute resources and proceed with the simulation on demand. In fact, the system is self-learning with a dynamic grow-in-time of knowledge to complement the databases. There are three methods of processing the data depending on whether one uses the knowledge base for rules-based prediction, or the patients' database for case-based prediction, or mathematical modelling for model-based prediction.

The system's final advice for treatment support is generated on the prediction of all three methods to produce the best solution. A few scenarios of disease evolution are produced and current treatment and a sequence of short-term and long-term treatments are being forecast by means of a decision tree, according to the speaker. Each modelled scenario has a rank of efficiency corresponding to the optimisation criteria chosen by the user. In line with this ranking, the system is able to select all scenarios for long-time forecast and for one step treatment.

The team of Dr. Sloot has used a relatively simple 2D cellular automaton to represent the spatial component of the immune system which allows to mimic the HIV dynamics in a correct way and test out drug therapies in various simulation models. The results are quantitatively corresponding to the clinical data. The prototype system used at the University of Amsterdam in clinical trials is called the Retrogram Decision Support System and has been set up together with Roche.

There are two ways of making the results available to the physician. The first one is the middleware approach in which the existing content of a Web site is being transformed to a variety of mobile devices supporting different operating systems, mark-up languages, micro-browsers and gateways. This is done by means of a proxy server. This proxy method has the advantage of excellently taking into account the limitations of mobile devices and wireless networks. However, it is a non-scalable solution whose information architecture is too dependent of the different devices and browser features. Syntactically, the result can easily be made acceptable but the challenge remains in making it user-friendly.

The second one is called the J2ME-HTTP and MIDlet approach. Here, the relevant information is grabbed and being shown in the MIDlets. The content is being retrieved from the original server into the device. It is a "write once, run everywhere" solution. Dr. Sloot defined the Java-approach as cost-efficient, portable, and providing a good user interface without any permanent proxy-device connection being required.

Dr. Peter Sloot ended his talk by stating that modern medicine constitutes an integrative science in which Grid technology will play an essential role because of its competitive infrastructure. The domain of infectious diseases forms an excellent test case for a Grid-based approach while both data and resources are distributed and roaming access is needed. From the security viewpoint however, a lot still needs to be done.

Leslie Versweyveld

[Medical IT News][Calendar][Virtual Medical Worlds Community][News on Advanced IT]