Determining the most efficient use of diagnostic tests is one of the complex issues facing the medical practitioners. It is generally accepted that excessive use of tests is common practice in medical diagnosis. Many tests are performed even though the incremental knowledge gained does not affect the course of diagnosis. With the soaring cost of healthcare in the US, there is a critical need for cutting costs of diagnostic tests, while achieving a higher level of diagnostic accuracy. Various decision making tools assisting physicians in diagnosis management have been presented to the literature. One such method, called analytical hierarchy process, utilize a multilevel structure of decision criterion for sequential pair wise comparison of available test choices. Many of the decision-analytic methods are based on Bayes' theory and decision trees. These methods use threshold treatment probabilities and performance characteristics of the tests, such as true-positive rate and false-positive rates, to choose among the available alternatives. Sequential testing approaches tend to elongate the diagnosis process, whereas the parallel testing approach generally involves higher number of tests. This research is focused on developing a machine learning based methodology for finding an efficient testing strategy for medical diagnosis. The method, based on the patient parameters (both observed and tested), recommends test(s) with the objective of optimizing a measure of performance for the diagnosis process. The performance measure is a combined cost of the testing, the risk and discomfort associated with the tests and the time taken to reach diagnosis. The performance measure also considers the diagnostic ability of the tests. The methodology is developed combining tools from the fields of data mining (rough set theory, in particular), utility theory, Markov decision processes (MDP), and reinforcement learning (RL). The rough set theory is used in extracting diagnostic information in the form of rules from the medical databases. Utility theory is used to bring three non-homogenous measures (cost of testing, risk and discomfort and diagnostic ability) into one cost based measure of performance. The MDP framework along with an RL algorithm facilitates obtaining efficient testing strategies. The methodology is implemented on a sample problem of diagnosing Solitary Pulmonary Nodule (SPN). The results obtained are compared with those from four other approaches. It is shown that the RL based methodology holds significant promise in improving the performance of diagnostic process.

Introduction : L'installation de caméras de Tomographie à Emission de Positons (TEP) a eu lieu en 2003 dans l'inter région Grand-Est de la France. Plusieurs études ont montré ses performances diagnostiques et ses bénéfices en évitant: des interventions de chirurgie exploratrice devenue inutile (thoracotomie aux séquelles douloureuses, laparotomie exploratrice), et certains examens complémentaires d'imagerie ou d'exploration fonctionnelle inutile. Objectifs : 1) évaluer le changement des stratégies diagnostiques du Nodule Pulmonaire Isolé (NPI) et du Cancer du Poumon Non à Petites Cellules (CPNPC) induit par l'implantation de la TEP dans l'inter - région Grand - Est de la France ; 2) mesurer l'impact du changement sur la qualité de vie des patients atteints d'un NPI et d'un CPNPC ; et 3) étudier le rôle pronostique de la QV sur la survie des patients atteint d'un CPNPC.Méthodes : Etude prospective multicentrique constituant 2 cohortes « Avant (2001 - 2002) » et « Après (2004 - 2005) » de l'implantation de la TEP. Ont été recueillis : les caractéristiques sociodémographiques des patients, clinique du NPI et du CPNPC, les examens diagnostiques et traitements réalisés jusqu'à 6 mois après la fin du bilan. Deux auto-questionnaires de QV (SF36 et QLQ C-30) ont été distribués à 3 et 6 mois après la fin du bilan diagnostique.Résultats : 1) Nous avons observé une diminution du nombre d'examens réalisés au cours du bilan diagnostique d'un NPI de 4 examens en phase avant à 3 en moyenne en phase après. Néanmoins, la fréquence des examens invasifs jugés évitables a posteriori pratiqués chez les patients ayant un NPI bénin n'était pas différente (47% phase avant versus 49% phase après). Un an après son installation, la TEP a été utilisée chez 11% des patients. 2) Les patients atteints d'un NPI et quelque soit sa nature, maligne ou bénigne, ont eu un score moyen de QV plus bas (-8 à -32 points, p<0.001) que celui mesuré dans la population générale française indépendamment du sexe et de l'âge. Les scores QV des patients ayant un NPI bénin se révélaient être statistiquement plus élevés que ceux des patients ayant un NPI malin, notamment dans les dimensions « fonctionnement social », « rôle physique » et « rôle émotionnel » (+10, +14 et +18 points respectivement, p=0,02 à 0,04).3) Les bons scores en fonctionnement physique en SF-36 (HR=0.78 ; IC=0.68 - 0.90 ; p<0.001), et en la limitation due à l?état physique en QLQ-C30 (HR=0.53 ; IC=0.59 - 0.89 ; p=0.003) indiquant une bonne perception des patients sur leurs fonctionnement physique sont liés à une meilleure survie. Les symptômes élevés de constipation (HR=1.18 ; IC=1.005-1.38 ; p=0.04) en QLQ-C30, par contre, est lié à une mauvaise survie. Le sexe féminin est lié à une meilleure survie (HR=0.55 ; IC=0,33 - 0,94 ; p=0,04). Cependant, le stade III et IV du CPNPC est associé à une mauvaise probabilité de survie (HR=1,72 ; IC=1,16 - 2,57 ; p=0,007).Discussion/conclusion : Les stratégies diagnostiques du NPI ont changé. L'impact de cette maladie, mesuré de façon subjective par la qualité de vie (QV) des patients peut indiquer certains domaines essentiels à améliorer dans sa prise en charge car une meilleure QV en oncologie broncho-pulmonaire est associée à une meilleure survie / Background: The Positron Emission Tomography was installed in 2003 in the North-eastern region of France. Previous studies showed its diagnostic performance and its benefices in avoiding: surgical exploratory interventions proved to be unnecessary a posteriori (for example, thoracotomy or laparotomy with painful consequences), and certain complementary imaging tests or unnecessary functional explorations. Objectives : 1) to evaluate changes in diagnostic strategies of Solitary Pulmonary Nodule (SPN) and Non Small Cell Lung Cancer (NSCLC) induced by the implantation of PET in the Northeastern region of France ; 2) to mesure the impact of these changes on the health related quality of life (HRQoL) of patients with SPN and NSCLC; and 3) to study the prognostic role of the HRQoL on the survival of patients with NSCLC.Methodes: An observational, prospective, multi-center design was applied. Two cohorts « Before (2001 - 2002) - After (2004 - 2005) » the PET implantation in the regions were settled up. Data collected included: patients' socio-demographic and clinical characteristics, the diagnostic tests and then treatments that were carried out during 6 months after the end of the diagnostic process. Two HRQoL questionnaires (the generic questionnaire SF-36 and the cancer specific one QLQ-C30) were distributed at 3 and 6 months after the end of the diagnostic process.Results:1) The number of diagnostic tests of patients with SPN decreased significantly from a mean of 4 in the before-PET to 3 tests in the after-PET period. Meanwhile, there was not any difference of the frequency of invasive tests considered unnecessary aposteriori for patients with benign SPN (47% before-PET versus 49% after-PET period). One year after its installation, the PET was used in 11 % of diagnostic strategies. In 7.7% of cases it was used after the bronchoscopy. 2) A comparison of the QoL with that of the French general population revealed that patients with SPN, whatever the diagnosis, benign or malignant, had worse mean scores (-8 to -32 points, p<0.001) compared to the general population with similar age and sex. A difference of 24, 30 and 32 points were observed in the "physical functioning", "emotional role" and "physical role" (p<0.001). The scores of patients with benign nodule were higher than those of malignant, especially on "social functioning", "physical role", and "emotional role" (+10, +14 and +18 points respectively, p=0.02 to 0.04). 3) Good scores on "physical functioning" of SF-36 (HR=0.78; CI=0.68 - 0.90; p<0.001) as well as "role functioning" of QLQ-C30 (HR=0.53; CI=0.59 - 0.89; p=0.003) were related to a better survival. Higher symptoms of constipation (HR=1.18; CI=1.005-1.38; p=0.04) in QLQ-C30 were associated to a worse one. While being a woman was associated to a better survival (HR=0.55 ; CI=0,33 - 0,94 ; p=0,04), the stage III and IV of the NSCLC was related to a lower one (HR=1,72 ; CI=1,16 ? 2,57 ; p=0,007) .Conclusion: The diagnostic strategies of SPN changed after PET was available for medical practice. This study showed the negative impact of SPN to patients' HRQoL. It indicates the domains in which health practitioners could interfere in order to improve the management of these diseases, because this study confirmed previous studies in pulmonary oncology, that patients' QoL is related to the survival