Optimized Healthcare Decision Support Method Based On Predictive Mining Techniques

The field of computing has undergone a drastic change. After years of relative stability, the world of Business Intelligence (BI) is now experiencing a sea change. BI is an indispensable component and an essential tool for understanding what drives an organization and ultimately leads to better decision making. BI tools exist at the heart of a Decision Support Systems (DSS) and are therefore not immune to the HealthCare sector. Data Warehouse is the foundation for Business Intelligence.The HealthCare environment is generally ’data rich’ but ’information poor’ and the Information Management system is on average very weak. The HealthCare sector is compounded with vast amounts of Clinical data which is seldomly used for decision making purposes. As storage density increases with patients’data, there is a great deal of attention in the HealthCare sector and an imminent need for turning such data ’tombs’ into useful ’golden nuggets’of information and knowledge. Surveys have proved that the Botswana Health Information Management System (HIMS) generally weak, fragmented, data collection techniques are not well coordinated, and duplication of information in data collection is very evident. This widening gap between data and information systems, calls for a systematic development of BI tools that will turn these ’data tombs’ into ’golden nuggets’ of knowledge.In this research, we propose to strengthen the Botswana HealthCare Information System, to optimize the DSS and Enhance Data Generation and Utilization for evidenced based decisions and planning. In order to achieve our main objective, we proposed to boost the HealthCare BI by optimizing its Decision Support Method based on Predictive Mining techniques. A quantitative, contextual design, co-relational, analytical and a conceptualized study are presented to strength the HealthCare BI. Our research objective was achieved through a synergy of building a basic Data-Warehouse, applying Data Mining (DM) techniques, a Data Analytical framework and a conceptual Business Intelligence Competence Center of excellence Framework. In carrying our study, we used the Anti Retroviral Therapy (ART) dataset obtained from the Ministry of Health (MOH) Botswana. Some HIV-Aids patients’under the ART programme continue to be at risk of contracting HIV/AIDS related Opportunist Infections(OI) and little evidence based research work has been carried out so as to apply preventative or mitigating factors for the patients who might be at risk. We aimed at building a Classification and prediction model by mining the historical data to determine those patients who might be at risk of contracting the afore-mentioned infections so that proper medical attention can be taken care of.Four classification and prediction models were built tested, analyzed for their robustness. The models were trained against the Generalized Linear Model (GLM), Support Vector Machine (SVM), Decision Tree (DT) and the Naive Bayes algorithms. We critically evaluated the performance of each model, analyzing the overall performance, using the Confusion Matrix analytical performance, the Receiver Operator Characteristic (ROC) curve and the LIFT Cumulative analysis. The SVM proved to be best model for our data with an overall performance of84%against the GLM Model which had a0%predictive confidence. The SVM had a total score of690correct predictions out of a total case count of820. The SVM also proved to have a higher likelihood of accurately predicting the negative or the positive class with a higher measuring impact at the60th quantile. This model had the highest measure of the Area under Curve of0.8941. Our OI SVM model was deployed to the rest of the data, and final the final prediction results were produced and published to an excel file format. The statistical analysis results proved that there was a significant correlation between Haemoglobin (HB), Viral Load (VL) and Weight of the study patients (p<0.05). This suggests that patients with higher values of body weight would also show increasing scores of HB and VL. However, there was no remarkable correlation between Age, BCD4and CD4.Our research work also presented a conceptualized BICC strategy of excellence framework which has proven that BI is no longer a one size fits all conglomerations of a few important single IT personnel. Basing on the outcomes that have been achieved after conducting this research work, there is still some interesting innovative aspects that are worthy to be worked on to improve the future of HealthCare BI. Lessons learnt from the limitations of this study can provide a platform for future innovative researches since technology is a revolving field.