Analysis of a process and computer user interface for data analytics results for patient discharge decision making

Initially, computer aids simply automated the processes that humans had developed to complete various tasks. Their primary contribution was to increase efficiency by automating repetitive, tedious tasks, both physical and cognitive. Many years later, sophisticated computer aids employ a broad-scale availability of large amounts of information, coupled with the ability to analyze that information using statistical methods. Data analytics can provide more information than ever before to those using computer based systems. Computer aids for engineering design are typically used by engineers, who are data-savvy, and are generally capable of using the results of data analytics directly incorporated into engineering design tools in an efficient manner. However, data analytics results are becoming more widespread and available for users who are not data-savvy. These systems must be designed with this in mind. This paper presents an example of a computer aid for healthcare, a tool to aid in the patient discharge decision. The motivating problem is the large number of patients who are readmitted to hospitals within 30 days of discharge. The design problem addressed here includes identifying the decision process the users should employ, as well as the computer/user interface that best uses the data analytics results. This work was carried out through collaboration with a healthcare provider. A readmission risk tool was developed using data analytics to estimate the probability of readmission based on a historical data set of 50 patient-specific factors. Preliminary work indicated that the users did not fully accept or utilize the data analytics results, and also demonstrated their need to continue to rely in part on their own expert heuristic decision processes. This study presents a method for dealing with these issues. First, a normative decision based approach to determining a multiattribute utility function is formulated and assessed to compare alternative computer/user interface designs. Then, the same approach is taken to assess the healthcare worker’s willingness to make tradeoffs under uncertainty when making the patient discharge decision. The resulting interactive computer interface design coupled with the normative healthcare decision process helps the user best exploit data analytics results while simultaneously facilitating the user’s continued deployment of their own expertise. Currently, the RRT is being utilized for flagging the readmission risk prone patients and not for making the main discharge decision regarding the patient. Its main use is to help the healthcare workers focus their attention on these medium-high risk patient profiles. Furthermore, the RRT result display needs to be made more actionable and relevant to the healthcare workers for the efficient use of the RRT model. We will discuss this in detail in our study