HA535 Data Analytics for Health Care Managers - Units 5 Descriptive Statistics - Visual Representations of Data

Question 1

What are analytics?

Answer 1

The systemic computational analysis of data or statistics.

More data than ever before are generated and available—particularly with the wide adoption of
EHRs.

The current regulatory environment requires the reporting of thousands of measures.

Hospitals and health systems are facing increased pressure to improve clinical, operational, and
financial results.

Population health has become a competitive strategy, and analytics is crucial to shaping
effective population health initiatives.

Information technology and software are increasingly sophisticated, allowing analysis of data
on a massive scale.

Question 2

Fun Fact: On average, the Centers for Medicare & Medicaid Services requires hospitals to report approximately on how quality measures for regulatory compliance?

Answer 2

1700.

The sheer number of data points that must be collected forces organizations to
dedicate significant resources to collecting and managing the data. And this effort does not take into
account the additional resources required to analyze and make decisions with the data.

Question 3

What is the goal of data analytics?

Answer 3

Obtain actionable insights that result in smarter decisions and better business outcomes.

Question 4

What is the basic work of an analyst?

Answer 4

Build a data framework using these major goals:

Gathering data—Data are the facts provided by databases

Building information—Information is the layer on top of data that helps make sense of the data.
Without essential knowledge of the business situation, the information is likely not valuable.

Gaining actionable insights—Actionable insights are those nuggets of knowledge from the
information that affect the organization. The insights should enhance a leader’s ability to make
improved decisions.

Question 5

What are the three distinct phases in which analytics take place?

Answer 5

Descriptive analytics
Predictive analytics
Prescriptive analytics

Question 6

What distinct phase in analytics is the process of condensing large data sets into meaningful information that can assist in decision making by examining past performance and summarization of data to discern trends and patterns to explain behavior?

Answer 6

Descriptive analytics.

In healthcare, reporting mechanisms such as regulatory compliance, quality measures, and financial results commonly use descriptive analytics.

Descriptive analytics makes up the largest subset of the analytics field. One main feature of
data visualization is making data consumable by people. The process of converting raw data is necessary because data alone are not typically usable to managers.

Question 7

What are some examples of descriptive analytics outputs?

Answer 7

Business intelligence reports (process of converting raw data through methods into info that help decision making)

Dashboards with key performance indications (KPIs)

Descriptive statistics

Traditional data visualization techniques

Question 8

What distinct phase in analytics builds models on the basis of data that can help forecast the future in terms of probabilities, provide insights for decisions, and it uses a variety of techniques ranging from regression modeling to machine learning to data mining to make projections?

Answer 8

Predictive analytics.

Question 9

Why have predictive models, a distinct phase of analytics, become popular in healthcare?

Answer 9

It is popular in disease management and population health.

For example, some healthcare organizations have begun to examine early indicators of diabetes to help prevent and lower costs associated with diabetes management (Barton 2016). This analytics activity is important as, according the Centers for Disease Control and Prevention (CDC 2009), more than 75 percent of total healthcare spending in the United States is related to chronic healthcare conditions.

At Hennepin County Medical Center (HCMC), the population health analysts discovered that
individuals diagnosed with HIV also suffered from poor nutrition. A predictive model was
constructed showing the positive impact of improved nutrition on healthcare costs.

Today, HCMC distributes healthy food with HIV medications for many of the patients in this population and have found overall costs to be reduced.

In short, predictive models have become a common approach to help reduce overall costs, improve quality outcomes, and lower overall patient risk.

Question 10

What are the three approaches used for developing predicative models?

Answer 10

Regressions, decision trees, and neural networks.

Regression-type approaches that can be used to predict future performance from historical data. Most analytical software (e.g., SAS, SPSS) packages include numerous regression tools.

Decision trees are a form of “supervised learning” tools. The decision tree algorithm first suggests a
split of the databases into a series of “leaves,” whereby each data point is allocated to one leaf. If the
analyst agrees with the computer’s selection of leaves, the computer then suggests a further
subdivision of the leaves. This process continues until the analyst believes the full tree represents a
good model of the data.

Neural networks attempt to mimic the human brain in the following ways:

Input units obtain the values of input variables and, if the analyst chooses, standardize those
values.

Hidden units perform internal computations, providing the nonlinearity that makes neural
networks powerful.

Output units compute predicted values and compare those predicted values with the values of
the target variables.

Question 11

Since regressions may be more accurate in predicative capability, what are decision trees more useful for?

Answer 11

Explaining the predictions to non analysts.

A version of the decision tree tool was used to create the Medicare diagnosis-related group (DRG) system in 1983. Exhibit 8.1 demonstrates the use of a decision tree to predict annual costs for Medicare patients.

Question 12

What distinct phase of analytics provides decision makers with models that offer guidance in form of recommendations which use a combination of predicative models, optimization, mathematical models, and other techniques?

Answer 12

Prescriptive analytics.

Examples of prescriptive models include the following:

Models for staffing that maximize quality outcomes and minimize costs
Models to maximize capacity in operating rooms
Strategic models that demonstrate efficient allocation of capital investments
Risk models that minimize adverse health events

Healthcare problems are complex and multidimensional and can be difficult to model. In the
modeling process, many assumptions are made in prescriptive models such as optimization.
Decision makers can use prescriptive models in combination with their knowledge of the healthcare
system to make effective decisions.

Question 13

What visualization of data helps users visualize the scale of differences between categories?

Answer 13

Bar or column graphs.

This is a classic example of a traditional business intelligence report created in Microsoft
Excel.

Question 14

What is another visualization of data that are useful in examining data over time?

Answer 14

Classic line graph, baby!

Exhibit 8.3 is a line graph showing the number of cases of biological
agents reported to the CDC from 1957 to 2012. The peak in the early 2000s represents the anthrax cases reported in the time frame following the 9/11 terrorist attacks on New York City and Washington, D.C., in 2001. As the exhibit demonstrates, line graphs reveal opportunities to explore trends and peaks in activity.

Question 15

What is a visualization of data where you create pictures to represent data where it might not necessarily help with prediction but demonstrates effectiveness in showing?

Answer 15

Map Functionality.

While such mapping does not have any predictive capability, exhibit 8.4
demonstrates its effectiveness in showing, for example, where the highest concentrations of reported diabetes patients reside. These types of maps help decision makers understand the concentration of data in geographic locations.

Question 16

What is a visualization of data where the relationships between two variables are shown?

What is a visualization of data that is a graphical representation of the distribution of data?

Answer 16

Scatter plots

Histograms

Question 17

What is a basic fundamental understanding of dashboards with KPIs (key performance indicators)?

Answer 17

In an analytics department, this is a technology is a data gathering tool with metrics and indicators that help showcase the progress towards objectives.

Question 18

What are the building blocks of many dashboard visualizations?

Answer 18

Metrics and KPIs.

In addition to being the products of an organization’s goals and objectives, metrics and KPIs may arise from strategy maps.

The definitions that follow build from one concept to the next and help inform dashboard
design. Take the time to understand each definition and the related concepts before moving on to the next definition.

Question 19

What, as a concept in dashboard visualizations, refers to a direct numerical measure that represents a piece of business data in relationship with one or more dimensions?

Answer 19

A metric.

Examples

Gross sales by week.

The measure is dollars (gross sales), and the dimension is time (week). For any given measure, viewing the values across different hierarchies in a dimension may be helpful.

For instance, a display of gross sales by day, week, and month shows the dollars (gross sales) measure along different hierarchies (day, week, and month) in the time dimension. The term grain refers to the association of a measure with a specific hierarchical level in a dimension.

Looking at a measure across more than one dimension, such as gross sales by territory and
time, is called multidimensional analysis.

Most dashboards do not leverage multidimensional analysis except in a limited and static way; more dynamic “slice and dice” tools are available in the business intelligence market.

This qualification is important to note. Say you uncover a significant
need for this type of analysis in the requirements gathering process. Knowing that these robust tools
exist, you have the option of supplementing your dashboards with some type of multidimensional
analysis tool.

Question 20

What is a key performance indicator (KPI)?

Answer 20

It is simply a metric that is tied to a target.

For example, our metric of gross sales by week with a goal of $10,000 would be our target.

Most often, a KPI represents the distance a metric is above or below a predetermined target. KPIs usually are shown as a ratio of actual to target and are designed to instantly let a business user know if he is on or off track without having to consciously focus on the metrics represented.

For instance, an organization may decide that, to hit the quarterly sales target, it needs to sell $10,000 worth of syringes per week. The metric is syringe sales per week, and the target is $10,000. Using a percentage gauge visualization to represent this KPI, and assuming we had sold $8,000 in syringes by Wednesday, the user would instantly see that he is at 80 percent of the goal.

When selecting targets for KPIs, remember that a target is needed for each grain you want to
view in a metric. Having a dashboard that displays a KPI for gross sales by day, week, and month,
for example, requires that targets be identified for each associated grain.

Question 21

What are the tools that can combine elements of each other but at a high level they all target distinct and separate levels of the business decision-making process?

Answer 21

Scorecards, dashboards, and reports.

Question 22

What tool is at the highest, most strategic level of the business decision-making spectrum and are primarily used to help align operational execution with business strategy with a goal to keep the business focused on a common strategic plan by monitoring real-world execution and mapping the results of that execution back to a specific strategy?

Answer 22

Scorecard.

Primary measurement used in a scorecard is the KPI.

These indicators are often a composite of several metrics or other KPIs that measure the organization’s ability to execute a strategic objective. One example of a scorecard KPI is profitable sales growth, which combines several weighted measures, such as new customer acquisition, sales volume, and gross profitability, into one final score.

Question 23

What tool resides one level below a scorecard in the decision-making process and is less focused on a strategic objective and more tied to operational goals which may contribute to one or more high-level strategic objectives?

Execution of the operational goal itself becomes the focus, not the high-level strategy.

Answer 23

Dashboards.

The purpose of a dashboard is to provide the user with actionable business information in a
format that is both intuitive and insightful. Dashboards leverage operational data primarily in the
form of metrics and KPIs.

Question 24

What tool is the most prevalent intelligence tool seen in business today which can be simple and static in nature, such as a list of sales transactions for a given time period, or more sophisticated cross-tab list of nested groupings, rolling summaries, and dynamic drill-through or linking?

Answer 24

Reports.

Reports are most appropriate when the user needs to look at raw data in an
easy-to-read format. When combined with scorecards and dashboards, reports allow users to analyze the specific data underlying their metrics and KPIs.

Question 25

When gathering KPIs and metric requirements for a dashboard, how should this be approached?

Answer 25

Traditional business intelligence projects often take a bottom-up approach in determining
requirements, where the focus is on the domain of data and the relationships that exist in those data.

When collecting metrics and KPIs for your dashboard project, however, taking a top-down approach
is preferred. A top-down approach starts with the business decisions that must be made first and then works down into the data needed to support those decisions.

To take a top-down approach, you must involve the business users who will be utilizing these dashboards, as these are the only people who can determine the relevancy of specific business data to their decision-making process.

Question 26

What is the vast majority of work being performed in healthcare analytics today?

Answer 26

Descriptive analytics mostly, with some highly specialized work in predicative and prescriptive analytics.

Almost all of these tasks share a common characteristic: They entertain a specific hypothesis. Examples are as follows:

I believe that patients of some doctors experience significantly longer lengths of stay than those
of other doctors for the same DRG.

I believe I can predict the amount of time a health plan will take to remit payment.

I believe I can predict which patients will not fill their prescriptions on the basis of their zip
code.

Question 27

What is a powerful approach, used in industries outside of healthcare, where data is explored without a specific hypothesis being established, relying only on a general sense that the data might reveal insights and is a subfield of computer science that uses algorithms to discover patterns of data interactions in large data sets?

Answer 27

Data mining.

It uses artificial intelligence machine learning, classical statistics, and advanced database systems such as Hadoop. Examples of data mining tools are clustering and text mining. Cognitive computing tools such as IBM’s Watson also support data mining.

Question 28

What places objects into groups suggested by the nature of data where they tend to be similar to each other in some sense, and these objects differ in different groups tend to be dissimilar?

Answer 28

Clustering, and clusters (instead of groups).

If obvious clusters or groupings are developed prior to the analysis, the clustering analysis can be performed by simply sorting the data.

After clustering is performed, the characteristics of the clusters can be examined graphically
using a clustering package in software such R or SAS statistical packages.

Exhibit 8.5 is a cluster analysis of the same Medicare data used for the decision tree in exhibit 8.1. Note that beneficiaries with chronic conditions cluster together because of their high use of inpatient services.

Question 29

In what situation would text mining be effective use of data gathering?

Answer 29

In EHRs, chart notes consist of many words.

Such as doctors’ and nurses’ notes. Therefore, a useful subset of data mining tools for healthcare providers is text miners. The case study that follows demonstrates the applicability of text mining to public health initiatives.

Question 30

For cognitive computing for data mining, how does it simplify work by utilizing a neural network?

Answer 30

Cognitive computing systems are designed to mimic human thought and provide natural language interfaces. A leading example is IBM Watson Analytics. Users load data into the system, and Watson performs significant preprocessing to suggest interesting correlations for the analyst to examine.

Exhibit 8.7 shows the starting screen from Watson as it looks at the Medicare beneficiary data
used in earlier examples. It immediately offers six questions for the analyst to pursue. It also
provides a natural language inquiry interface to delve deeper into the data.

Watson is a sophisticated example of a supervised learning tool and will continue to evolve as
its underlying artificial intelligence software improves.