Healthcare Data Analytics

Data analytics-based approach is the new frontier in improving the quality of healthcare while reducing cost. Studies have shown that data analytics are transforming the healthcare marketplace to promote a cost-efficient interactive healthcare system that provides better patient outcomes. For example, data analytics and a unified data management system integrating multiple sources of data including electronic health records, claims data, test results, and questionnaire data enable a healthcare provider to deliver a comprehensive, actionable picture of what's going on with each individual patient. A care provider, using the data, can predict which patient is likely to develop an acute condition over the course of a year, and can then be paired up with a doctor who is instructed to do whatever is necessary to diagnose the patient and prevent a major and expensive healthcare crisis. Data from healthcare fields are incredibly rich including electronic health records, biomedical image, sensor data, biomedical signal, genomic data, clinical text, biomedical literature, and social media data. Such a wide variety of data require physicians, nurses, and other healthcare professionals to be proficient in modern data analytics, or at the very least be exposed to developments in this emerging frontier of healthcare.

Analytics starts with the collection, organization, and manipulation of data for decision making, and includes three major components:

1. Descriptive Analytics categorizes, characterizes, consolidates, and classifies data to convert it into useful information. Two important areas of descriptive analytics are visual analytics and lean six sigma. Visual Analytics is emerging as an important field in healthcare given the rapidly increasing health-related digital, optical, and graphical information. Visual analytics combine human cognition, interactive visual interfaces, and data analytics to facilitate interpretation of complex data and gain insights. For example, clinical data are considered lumpy (consisting of immense amounts of patient demographic data), noisy (relevant or important signals often hidden or clouded by other random factors), and temporal (change constantly with time). These factors make it difficult for users to synthesize the information and obtain insights from the data. New interactive interfaces are needed to explore these kinds of clinical data. Visualization tools such as SAS Visual Analytics, JMP, and Tableau are becoming more capable and can be customized to explore clinical data in more depth. Lean Six Sigma focuses on eliminating defects and improving the patient experience by making sure processes consistently deliver the desired results. Graphical tools are an essential component of this methodology.
2. Predictive Analytics is about predicting the future by examining historical data, detecting patterns or relationships in these data, and then extrapolating these relationships forward in time. Data from healthcare fields are incredibly rich including electronic health records, biomedical imaging, sensor data, biomedical signal, genomic data, clinical text, biomedical literature, and social media data. Such a wide variety of data require modern data analytics. With analytics, one can build clinical prediction models for diagnosis and treatment of diseases. Common examples in Healthcare include:

(a) Sensor data such as physiological data (e.g. ECG, EEG) are obtained by monitoring patients on a regular basis and are measured real time at high frequency. Signal processing and time-series analysis techniques are applied to such data in order to isolate the relevant from such an immense collection of information.

(b) Clinical notes form the backbone of healthcare data. Natural Language Processing (NLP) methods such as shallow or full parsing, textual analysis, sentiment analysis, and other data mining techniques can be applied to extract information from the clinical text and to detect errors.

(c) Stream analytics can allow us to mine sensor data and biomedical signals.

(d) Social media analytics and Geo-spatial analysis/Geographic Information System (GIS) will help detect health trends such as outbreaks of infectious diseases, detect adverse drug interactions, and improve interventional capabilities for health-related activities, and analyze customer satisfaction.

(e) Sophisticated analytics methods have been developed to detect healthcare fraud, waste, and abuse (FWA).

Here are some common topics in Predictive Analytics: classification (logistic regression, discriminant analysis, cluster analysis); text mining, natural language processing; sentiment analysis; web analytics/web mining; search engine optimization; social network analysis (Facebook, Linkedin, Twitter); time series forecasting (ARIMA, ARCH/GARCH, ); sample survey; marketing Research; marketing analytics; conjoint analysis; econometrics; regression; principal component analysis; factor analysis.

3. Prescriptive Analytics uses optimization to identify the best alternatives to minimize or maximize some objective. Many circumstances in healthcare require the implementation of decisions that are made hoping to optimize costs and efficiency while still delivering the best in patient care. Making these "optimal" decisions can no longer be left to reason and human analysis alone; to optimize decisions, prescriptive analytics renders complex information clear by essentially constructing models of all possible outcomes and providing best case scenarios. Below are two applications of prescriptive analytics in Healthcare:

(a) Cost adjustments: When payers introduce high co-payments to limit the use of expensive drugs, costs may balloon elsewhere in the system because patients overall health can deteriorate without the drug - subsequently requiring even more services at higher costs. Such "penny wise and pound foolish" outcomes can be captured and appreciated only if the healthcare professional has training and adequate insights in effective optimization strategies and techniques using data from the field.

(b) Medical facility layout: Determining the optimal number of rooms needed for emergency departments, physician offices and surgical departments can no longer be left to elementary patient flow data. More complex analysis considers a myriad of factors before making these important decisions. A successful local example of this type of analytics utilized a Monte Carlo simulation at the UC Davis Medical Center and optimized the number of operating rooms needed while minimizing patient wait times and managing costs. Here are some common topics in prescriptive analytics: linear programming, nonlinear programming, integer programming, decision trees, genetic algorithms, Monte Carlo simulation, discrete event simulation, visual interactive simulation.

4. A closely related area is Machine Learning. Machine Learning for Healthcare involves developing algorithms that learn to recognize complex patterns within rich and massive data. Challenging problems such as clinical genomic analysis and the design of clinical decision support systems require applications of machine learning. Machine learning can be also used for more mundane situations. For example, mobile phones and facial recognition software can be used to determine if the right person is taking a given drug at the right time. The same technology can capture patient data and automated algorithms can be employed to track patient progression and to identify needed intervention through recognition of complex patterns from massive data. These algorithms can also detect or flag adverse events in a timely manner alerting the clinician. Here are some common topics in machine learning: recommender systems (recommendations at Amazon, Facebook, Linkedin), collaborative filtering, association rules, support vector machines, neural networks, naive Bayes classifiers, nearest neighbor method, classification trees, random forests.
5. There are several other interesting developments related to Big Data: Hadoop, MapReduce, Stream analytics (data in-motion analytics, real-time data analytics), Internet of Things (IoT) Analytics, Geo-spatial analytics and GIS.