Neural networks analysis of spontaneous pneumothorax development

Spontaneous pneumothoraces (SP) tend to cluster. Correlations between SP and atmospheric variations were reported by previous studies. In our work SP correlation with meteo variables and air pollutants in Cuneo County was analyzed. 2004-2010, 451 SP patients were prospectively evaluated. For each day of analyzed period, meteo parameters and pollutants were recorded. Statistics on SP evaluated distribution characteristics, spectral autocorrelation and spectral analysis; multivariate regression techniques were performed using artificial neural networks. Analysis of seasonal distributions showed no significant correlation. Spectral analysis showed that SP events were not random. Correlations between meteo-environmental variables were analyzed through linear tests. Neural networks showed some variables may predict SP insurgence. SP occurrence significantly increases in warm windy days with high atmospheric pressure and high NO 2 concentration. These data don9t affect SP treatment; nevertheless, they add information on SP tendency to cluster.