Comparison of artificial neural networks and logistic regression for determination of social-cognitive factors impacting drug abstinence

ABSTRACT

Background: Logistic regression model (LRM) are popular in public health studies. However, most studies have found that artificial neural networks (ANNs) have better performance than other methods especially when the relationship between the response and independent variables is complex.

We aimed to compare the performance of LRM and ANNs in the determination of drug abstinence predictors.

Methods: The participants of this prospective study included 317 male addicts recruited through convenience sampling in five short-term residential substance use treatment programs in Southern Khorasan, the eastern province of Iran. Eligibility criteria included being in a detoxification stage and having a primary diagnosis of substance dependency disorder based on the diagnostic criteria of the DSM-IV-TR. The performance of the models was compared using the area under the ROC (AUC).

Results: LRM and ANNs method correctly classified 82.2% and 84.4% of the participants as an abstinent group, respectively. The areas under ROC curve were 90 and 91.2, the kappa statistic was 0.643 and 0.688 for LRM and ANN models, respectively.

Conclusions: The results showed classification performance was not significantly different. We concluded that the selection of prediction techniques for predicting drug abstinence factors depended on our background knowledge on the data pattern.

KEYWORDS:

• Neural networks
• logistic regression
• abstinence

Author statements

All procedures were approved by the Ethics Committee of Isfahan University of Medical Sciences (Code No. 393198). No funding was received for this paper.

Ethics

All patients participated voluntarily after they had been fully informed and obtain written informed consent.

Disclosure of potential conflicts of interest

The authors have no conflict of interest to report.