Naslov
Applicaton of data mining methods in diagnosing posttraumatic stress disorder

Autori
Kozarić-Kovačić, D. ; Marinić, I. ; Jendričko, T. ; Gamberger, D. ; Supek, F. ; Kovačić, Z. ; Rukavina, L.

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
Neurologia Croatica, Book of abstracts, 56 (Suppl. 2): 1-144 / Petravić, D: - Zagreb, 2007, 93-93

Skup
The Second Croatian Congress of Neuroscience

Mjesto i datum
Zagreb, Hrvatska, 18.05.2007. - 19.05.2007

Vrsta sudjelovanja
Poster

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
data mining server; diagnosis; PTSD; posttraumatic stress disorder

Sažetak
Introduction. Dana mining is a term that refers to discovernig information and knowledge from existing data ; one important condition is the existence of uniform data sets. The aim of our study was to explore the possibilities of data minig methods in creating a diagnostic algorithm for posttraumatic stress disorder (PTSD). Materials and Methods. Using Random Forest classifier and Data minig server intelligent data analysis intelligent data analysis methods we have built several models for predicting a PTSD diagnosis, based on data from 102 inpatients (51 with diagnosis of PTSD and 51 with psychiatric diagnosis other than PTSD). The models were based on data from the structured psychiatric interview, psychiatric scales (Clinician-Administered PTSD Scale (CAPS), Positive and Nagative Syndrom Scale (PANSS), Hamilton Anxiety Scale (HAMA), Hamilton Depression Scale (HAMD)) and combined data from these sources. Results. The model based exclusively on the structured psychiatric interview distunguished comorbid diagnoses (neurotic, stress-related and somatoform disorders) from PTSD. Using Data minig server, a diagnostic algorithm was constructed with moderate sensitivity and specificity. In the second model along the same group of diagnoses, the elements of the CAPS score and PANSS additional criteria score were found to be importaint. A diagnostic algorithm was made, using Data mining server, with a higher sensitivity and specificity than the previous model. In the third model, the attributes from psychiatric scales were found to be most important., together with comorbid diagnoses of neurotic, stress-related and somatoform disorders. Conclusion. Some importaint attributes from each of the models were recognized that could be used in a diagnosic algorithm and some preliminary models for diagnosing PTSD have been built. Our work demonstrates one of the possible applications of complex data mining methods in creating diagnostic algorithms for clinical use.

Izvorni jezik
Engleski

Znanstvena područja
Kliničke medicinske znanosti

POVEZANOST RADA