A Systematic Review on Healthcare Analytics: Application and Theoretical Perspective of Data Mining

Tags

Top Highlights

• Healthcare is a booming sector of the economy in many countries [1]. With its growth, come challenges including rising costs, inefficiencies, poor quality, and increasing complexity [2]. U.S. healthcare expenditures increased by 123% between 2010 and 2015—from $2.6 trillion to $3.2 trillion [3].

• Inefficient—non-value added tasks (e.g., readmissions, inappropriate use of antibiotics, and fraud)—constitutes 21–47% of this enormous expenditure [4].

• Some of these costs were associated with low quality care—researchers found that approximately 251,454 patients in the U.S. die each year due to medical errors [5].

• Analytics provides tools and techniques to extract information from this complex and voluminous data [2] and translate it into information to assist decision-making in healthcare.

• It can generate fact-based decisions for “planning, management, measurement, and learning” purposes [2]. For instance, the Centers for Medicare and Medicaid Services (CMS) used analytics to reduce hospital readmission rates and avert $115 million in fraudulent payment [8].

• There is no comprehensive review available which presents the complete picture of data mining application in the healthcare industry.

• These studies are also limited in the rigor of their methodology except for four articles [11,16,22,25]

• Figure 4

• Major drawbacks were the absence of data source and performance measure of data mining algorithms.

• Figure 8 Utilization of data mining techniques, (a) by percentage and (b) by application area.

• For example, Yeh et al. [38] developed discrete particle swarm optimization based classification algorithm to classify breast cancer patients from a pool of general population.

• We recommend applying multiple algorithms and choosing the one which achieves the best accuracy.

• Figure 10 Percentage of papers utilized healthcare analytics by application area (92 articles out of 117).

• Discrete event simulation (i.e., modeling system operations with sequence of isolated events) is a useful tool to understand and improve ED operations by simulating the behavior and performance of EDs.

• Previously, the “casemix” principle, which was developed by expert clinicians to groups of similar patients in case-specific settings (e.g., telemetry or nephrology units), was used, but it has limitations in the ED setting [58]

• Researchers applied [58] data mining (clustering) to the ED setting to group the patients based on treatment pattern (e.g., full ward test, head injury observation, ECG, blood glucose, CT scan, X-ray).

• Using a large amount of ICU patient data (specifically from the first 24 h of the stay) collected from University of Kentucky Hospital from 1998 to 2007 (38,474 admissions), one group of researchers identified 15 out of 40 significant features using Pearson’s Chi-square test (for categorical variables) and Student-t test (for continuous variable) [59].

• Integration in healthcare system Very few articles reviewed made an effort to integrate the data mining process into the actual decision-making framework. The impact of knowledge discovery through data mining on healthcare professional’s workload and time is unclear. Future studies should consider the integration of the developed system and explore the effect on work environments.

• For example, a set of problems and medications can co-occur frequently. If a clinician has knowledge about this relation, he/she can prescribe similar medications when faced with a similar set of problems.

• Pre-processing of data, including handling missing data, is the most time-consuming and costly part of data mining.