MHI 584: Chapter 5

Chapter 5 Notes

Big Data

large databases

in healthcare they contain considerable information on behaviors, individuals, or small population sectors

List the 4 types of knowledge-based IT systems that inform and transform clinical decision making

EMR

HIE

EHR

PHR

*each supports a level 1 analysis

data mining

use of sophisticated search capabilities and analytical techniques on large data bases to discover patterns, correlations, and trends that can be leveraged to produce knowledge

uses mathematical algorithms to convert the accumulated experiential knowledge embedded in data files into explicit knowledge

What constitutes a large database?

the application and types of analytics deployed

What is the size of the database defined by?

the statistical tool and level of analysis being performed

What are the 2 main components of data mining?

stored data

mathematical algorithms

stored data

the various databases that are available

has a great impact on the nature of data-mining methods used

mathematical algorithms

comprise a set of precise rules followed by the computer in calculating relationships in a database used for knowledge extraction

List the 5 main types of databases

prerelational

relational

object oriented

resource descriptive format (RDF)

nonstructured query language (NoSQL)

prerelational database

stores data in tree-like hierarchies

relational database

keeps data in tables tat represent real objects connected through relationships

the most widespread database format used in some EHRs

its structure is powerful and adaptable for CDSS but somewhat inflexible

object oriented database

more flexible than relational databases

can easily deal with a variety of objects

integrate data with code under the object-oriented paradigm resulting in more structured representation of the problem domain

introduces the concept of class hierarchies which allow for incremental refinement of the domain model

resource descriptive format (RDF) database

addresses most of the problems inherent in other databases types

stores data in one table with 3 columns: subject, predicate, and object

all information is stored as triples

has an extensible format that is easy to interconnect and conducive to mining

require a way to “chunk” information into triples through NLP

nonstructured query language (NoSQL) database

doesn’t store data in a set structure (tables) and uses key-value mechanism for data retrieval instead of SQL query

are scalable and more suitable to diverse content than relational databases are

lack transactional support so they are less suitable for health data

inflexibility

a problem with IT architecture that was designed for a given purpose because if the problem context changes modifying the tables to reflect the new situation is not easy

List reasons why relational databases are not conducive to data mining

format minimizes data redundancy and is suitable for multiple input-output operations resulting in users having to search many tables to retrieve the desired data

format doesn’t allow the discovery of relationships among data that were not already known when the database was built

data warehouse

data from a broad range of sources linked together and stored for easy retrieval, reporting, analysis, and decision making

highly optimized for output (fast data retrieval) but not for input

online analytical processing (OLAP) architecture

IT setup that enables the user to “slice and dice” the data in multiple dimensions to provide insights

highly optimized for fast data input-output or for satisfying the demand for medical personell during the care delivery process

natural language processing (NLP)

computational approach to processing human language

increasingly employed in contemporary EHRs to enable searching of unstructured text fields

contribute to the transition from EHR systems to data mining and analytics knowledge management systems

List the historical evolution of information systems

file based (i.e. single desktop computer)

relational & object oriented (i.e. desktop computer and local network)

RDF (i.e. desktop computer and internet)

NoSQL

What distinguishes data-mining methods from hypothesis-driven data analysis?

data mining generates rather than verifies hypotheses

What type of relationship do knowledge and data have with respect to the data-to-information-to-knowledge-to-wisdom paradigm?

feedback relationship as we need to know something about problems before data mining

Why is the term data mining considered restrictive?

it suggests that the analysis part of data mining is performed after a large quantity of data has been accumulated

*data mining includes statistical analysis

List the 3 main types of data-mining methods

classification

clustering

association-rule mining

classification

an algorithm that attempts to assign an unknown object to one of the available classes of known objects

can be data driven or knowledge driven

clustering

the quintessential data-mining problem

the number of groups and the relationships between the objects are typically not known

association-rule mining

another typical data-mining algorithm

tries to find relationships among characteristics of objects stored in a database

discovery is based on the frequency of association (if two characteristics associate often, then their relation might be relevant to the problem being analyzed)

data-driven algorithms

previously stored data are used together with the known class labels to develop classification models

i.e. neural networks, support vector machines, or simple Bayes ones

knowledge-driven algorithms

enterprise

refers to the system & includes all units that bring knowledge from organizational and financial to the services being delivered

extends beyond individual clinicians, the organization, and the health system to include other sectors such as education and social services

What defines the structure of the enterprise?

the nature of the problem being addressed and the collective knowledge need to serve the consumers

What does knowledge management include?

considering how the clinical and business enterprise might be structured to bring maximum knowledge to bear on a problem being considered

modeling

method of studying, understanding, and then replicating the complexities of the real world in order to design, change, and improve systems

starts with building conceptual models that challenge current assumptions and introduces futuristic thinking

transforms the system structure making it the independent variable

a means of increasing knowledge for both clinicians and organizational leaders

analytical models

can be developed and applied to test assumptions, refine the conceptual model, and present alternative futures

dynamic modeling

allows transformation to be accomplished through complex analytical models, changing financing, organizational design, and other system components until the system achieves an optimal state

What do organizational leaders need to effectively develop new models

tools that frame the issues and test alternative assumptions

building conceptual mental models

a process of finding a common framework

draws on the perspectives of each participant but also challenges and extends these perspectives

defender

relies on existing strategies

reactor

reacts to others

prospector

innovates

complexity leadership

characterized by leadership teams that interact within and across organizational domains and that emphasize both informal and positional leaders

Detail the modeling process

information from the real world provides some understanding that forms the paradigm or mental model of decision makers

decision makers then use paradigm to build a model

the model along with additional information from the real world (e.g., through data mining), brings about a paradigm shift

process is repeated until model sufficiently represents real world and decision maker gains knowledge or learns along the way

model is then used to predict future states of the system and to evaluate what-if scenarios

information & knowledge is then used to act in the real world

models

can be statistics based, optimization based or simulation based

used to study static and dynamic systems

are all “wrong” from an idealistic standpoint as they cannot represent each and every aspect of reality

represent the key elements of a system, frame key questions, and propose approaches to solutions

help challenge and break down mental models/silo thinking and prompts paradigm shifts

Why is the validity of models important?

to ensure they succinctly and sufficiently capture the real-world system

discrete event (DE) modeling

used primarily to study processes, streamline them, and reduce bottlenecks through better resource allocation, capacity utilization or standardization, and mechanization of routine processes

used to improve processes within but not to restructure healthcare organizations and systems

typically focuses on operations (including processes that transcend professional and institutional boundaries) within a fixed, stated, or assumed strategy

requires understanding queuing theory

decomposed systems

the processes within healthcare organizations and systems that have the greatest potential value

List skills required for performing DE modeling

flowchart or process mapping

data collection

fitting arrival and service distributions

model building in simulation software

quantitative analysis for staffing or capacity planning

agent-based (AB) modeling

used to study the behavior of systems on the basis of the interactions among agents or entities

uses the behavior of individual agents under given circumstances to model the overall changes in the system over time

related to DE modeling/bottom-up approach where the agent is typically an individual or a functional area of an organization

its strength is its interdisciplinary nature as it can synthesize knowledge form different disciplines

List skills required to perform AB modeling

developing state charts

modeling interactions

collecting data

building models in simulation software

performing scenario analysis

health systems informatics

addresses the complexity of sectors and systems to enable leaders to develop the most complete understanding of the interactions among disparate systems

systems dynamic (SD) modeling

used to model complex, nonlinear relationships between components and to study the dynamics of the system over time

framework operates under the premise that structure predicts behavior over time

complexity theory is the underlying principle

complexity theory

views the organization as a learning system that uses knowledge to drive the organization’s strategies and structures

challenges the school of thought that promotes prescriptive structures, plans, and strategies

List the most relevant skills for SD modeling

systems thinking

cause-and-effect formulation

data collection

stock-and-flow modeling

differential and integral calculus

model building in simulation software

analysis for decision making and policymaking

Why has the value given to high-order systems applications been low?

because of the highly regulated and subsidized nature of the healthcare industry, the power held by health professionals, and healthcare leaders’ traditional orientation toward the business function

How do the 3 modeling approaches compare

their level of abstraction and the amount of detail required for modeling (DE modeling is at the middle to low level of abstraction, but it requires a medium to high amount of detail for modeling purposes while SD modeling is at the high abstraction level and is used for strategic decision making and policy analysis & AB modeling spans a wider range of abstraction, including high, middle, and low levels of abstraction)

DE and AB modeling are predominantly discrete, whereas SD modeling is continuous

DE and SD modeling can be used together to study the nuances of operations and to learn the impact of strategic decisions

When is an information system considered optimized?

when it contains the collective knowledge of all agents, who engage in dialogue to define the problem and develop an optimal solution strategy