ANFIS PREDICTION OF OUTPATIENT’S NON-ADHERENCE TO MEDICATION

ABSTRACT

Leveraging machine learning algorithms to accurately predict patients who are unlikely to adhere to their prescribed medication and to target them with delivery of personalized and persuasive messages can be very effective and efficient in improving medication adherence. In this study, ANFIS prediction of patient’s non-adherence to medication and intervention system was developed to improve medication adherence. Over the years, several systems have been developed to address the multifactorial problem of non-adherence among outpatients. However, these systems have been inefficient and sub-optimal in performance due to inaccurate identification of non-adherent patients and failure to detect the causes of medication non-adherence in individual patient. A blend of survey, correlation and experimental research method was adopted in this study due to its multidisciplinary nature. Outpatients’ non-clinical dataset of 609 records was generated through a validated questionnaire-based survey administered at three tertiary healthcare centres in the South East Region of Nigeria. Four features with highest impact of sensitivity and correlation with non-adherence feature were selected from the dataset which are patient’s behavioural pattern, knowledge, perception and belief as input variables. The ANFIS predictive model performed well with the root mean square error (RMSE) being 0.391 and 0.416 in training and testing respectively. It was able to predict and stratify non-adherence level with high significant accuracy. The new system has the capability to greatly improve medication adherence as it used ANFIS predicted outcome to drive adherence intervention activities.

Keywords: ANFIS, medication non-adherence, prediction, machine learning algorithm, outpatient