The Decision to Adopt Machine Learning for Telemedicine

Telemedicine is fast-growing as a mobile health care information system (HIS) in most parts of the world. Fast Internet, smart phones and increased comfort of physicians in using electronic communication are also helping telemedicine become more widely adopted. Telemedicine consultation can contribute to reducing cost, lessening the stress of patients and improving accessibility to specialized consultations. However, it is difficult to schedule correct telemedicine sessions without a deep understanding of the health care needs of the region. The use of machine learning for decision making and better treatment has been a highly researched topic. Machine learning is also used to monitor patients remotely. However, this technique is not currently used to monitor telemedicine session broadcasting. In our recent Journal article, we present the case of an Indian health care organization that broadcasts telemedicine sessions to associated hospitals in remote locations. For the purpose of telemedicine governance, we suggest the following steps while using machine learning techniques through the department-session-organization (DSO) model proposed in our article:

• Understand the specific IT governance problem using organization mission and vision to determine the purpose of the prediction model.
• Past data collection, data cleaning to remove incomplete data and analysis of the data is required.
• Perform data transformation for simplification and improved decision making if needed. For example, we simplified our model by clustering hospitals based on regions and identified teaching and nonteaching hospitals for better distinction and prediction.
• Based on the data set, the organization needs to determine the kind of machine learning technique suitable for its decision making. In our study, as the variables were categorical and best suited for a classification model, we tested multiple classification techniques. Based on the results, we observed that a classification tree provided us the best prediction accuracy.

It is also important to balance the cost of information retrieval and resulting profit out of the prediction technique. While determining the return on the additional investment, we accounted for the risk associated with misclassification by the telemedicine decision support system (TDSS). A clear understanding of the risk and return on investment will help the hospital to understand the pros and cons of going forward with such a prediction technique.

Read Shounak Pal and Arunabha Mukhopadhyay’s recent Journal article:
“A Machine Learning Approach for Telemedicine Governance,” ISACA Journal, volume 1, 2017.

Shounak Pal and Arunabha Mukhopadhyay, Ph.D.

[ISACA Journal Author Blog]