TERÀP-IA sistema expert d'ajuda al tractament de les pneumònies

Es presenta un sistema expert (SE) que es justifica per: 1. Saplica al tractament duna malaltia frequent. 2. Saplica a un domini on hi ha incertesa. 3. Pot integrar tota la informacio necessaria per prendre una decisio adequada, i pot disminuir la variabilitat de la prescripcio. 4. Pot fer reproduible la decisio, ja que els criteris que utilitza per prendre-la son reproduibles i es poden explicar. Els objectius de la tesi han estat: 1. La modelitzacio del coneixement medic necessari per desenvolupar TERAP-IA. 2. La generalitzacio del model conceptual al tractament daltres infeccions. 3. La implementacio del SE. 4. La seva validacio. Quant al 1er objectiu, lanalisi del problema ha permes identificar el coneixement farmacologic (CF) dels antibiotics que son actius per tractar les pneumonies i el coneixement de les dades del pacient (DP) que son necessaries per deduir el millor tractament. Tambe ha permes descriure unes tasques que, realitzades successivament, ajuden a triar el tractament mes adequat. Aquest coneixement i aquestes tasques shan representat en una arquitectura que sha generalitzat al tractament daltres infeccions, de la qual TERAP-IA es una particularitzacio. TERAP-IA sha implementat amb MILORD II llenguatge basat en moduls. Els moduls de TERAP-IA son: adquisicio de DP; moduls de CF; moduls de filtratge dels antibiotics per algunes condicions del malalt, ex.: lembaras, que modifiquen ladequacio dels antibiotics disminuint el seu valor o eliminant-ne alguns; moduls de microorganismes (MM) que proposen un tractament antibiotic per a cada un dels microorganismes que pot causar pneumonia; modul de Combinacions que combina els resultats dels MM, i proposa el tractament de conjunts de microorganismes; filtratges de les combinacions que trien el millor tractament segons la dosi, espectre i preu. En aquests moduls els conceptes relatius al CF i les DP shan representat amb fets. Per deduir nous fets shan utilitzat regles. Per expressar la incertesa sha utilitzat una logica proposicional multivaluada basada en un conjunt detiquetes linguistiques entre segur i gens possible. Els resultats de TERAP-IA son antibiotics amb un valor de certesa. La validacio dels resultats del SE, al no disposar d'un patro de referencia, ha consistit a comparar les respostes de SE amb les de 5 metges experts (ME) respecte a 58 histories cliniques de pacients amb pneumonia i classificar els resultats del SE en relacio als resultats dels ME mitjancant diferents mesures. El conjunt de tractaments proposats pels ME i pel SE s'ha estructurat en una matriu de dades a partir de la qual shan obtingut 3 matrius de distancia: euclidiana, city-block i Mahalanobis i una matriu de concordanca: lindex Kappa ponderat. A aquestes matrius sels hi ha aplicat un analisi de clusters que mostra que el SE no s'aparta de l'acord que tenen els ME entre ells mateixos. An expert system (ES) is presented. The justifications of its use are: 1. It is useful for the decision-making therapeutic process of a frequent disease. 1. It is applied in a domain with uncertainty. 2. It can integrate all the necessary information for a good decision, and can decrease the variability of the prescription. 3. It makes the decision reproducible, as the used criteria are reproducible, and can be explained. The aims of the thesis have been: 1. The modelization of the medical knowledge necessary for the development of TERAP-IA. 2. The extension of this conceptual model to the treatment of other infectious diseases (ID). 3. The implementation of the ES. 4. Its validation. In relation to aim 1 the analysis of the problem has allowed to identify the pharmacologic knowledge (PK) of the antibiotics useful for adult community acquired pneumonia (CAP) treatment, and the knowledge of all the patient conditions (PC) necessary for the choice of the antibiotic therapy. It has also allowed to describe a group of procedures that, performed successively, are very helpful for the best therapeutic decision. These knowledge and procedures have been represented in a construct that has been extended to the general treatment of ID, being TERAP-IA a particularisation of it. TERAP-IA has been made operational by using MILORD II, a modular language for knowledge-based systems. TERAP-IA modules are: 1. Acquisition PC. 2. PK modules. 3. Modules developed for refining decision of therapy according to specific patient conditions, i.e. pregnancy. 4. Modules proposing a treatment for every one of the micro-organisms involved in CAP. 5. Module that combine the results of the micro-organism modules to obtain the treatment of set of micro-organisms. 6. Filter modules of the best treatment, taking into account the dosage, price and action spectrum of drugs. In these modules the concepts have been represented as facts. For the deduction rules have been used. A propositional many-valued logic based on a set of linguistic terms from sure to impossible was used to express the uncertainty. The results of TERAP-IA modules are antibiotic treatments with a truth-value. As we did not have a gold standard for comparing the results of ES, the validation was done by comparing the results of the ES with those of the 5 expert doctors (ED) in relation to 58 clinical records of patients suffering of CAP. The ES results were classified in relation to the ED. For this purpose different measurements was used. All the therapeutic possibilities for every case were included in a data matrix, from witch 3 distance matrices were obtained: euclidian, city-block, Mahalanobis, and a concordance matrix: the pondered Kappa index. A cluster analysis was applied to these matrices, indicating that in no one of the measurements analysed a difference could be observed between the ES and the agreement of the ED.