Business Intelligence Best Practices: A Strong Foundation ... · Other Categories (18) 42...
Transcript of Business Intelligence Best Practices: A Strong Foundation ... · Other Categories (18) 42...
1
Business Intelligence Best Practices: A Strong Foundation for Organizational Success
WSB, February 19, 2017
Joseph C Nichols MD, Health Data Consulting
Thomas Harlan, Iatric Systems
2
Speaker IntroductionJoe Nichols, MD - Principal
Health Data Consulting Inc
• >35 years in the healthcare industry
• 15 years in private orthopedic practice
• 18 years healthcare IT
• National speaker, trainer and consultant for CMS, Vendors, Associations, Hospitals, Payers and Professional Practices. Over 200 presentations nationally on health data and coding
• AHIMA approved ICD-10 coding trainer
• Three service awards from WEDI (Workgroup for Electronic Data Interchange)
• Member; board of directors of the University of Washington Health Information Management masters degree program
• Member of the HIMSS Revenue Cycle Task Force
3
Conflict of Interest
Joseph Nichols, MD
Has no real or apparent conflicts of interest to report.
4
Agenda
The changing role of healthcare data and analytics in a value-based purchasing environment
Historical data quality challenges
The challenge of data aggregation
The impact of patient diagnostic data
Re-focusing analytic efforts toward value based purchasing
5
Learning Objectives
Describe the change in healthcare policy and payment focus that is driving new analytic requirements
Explain the current challenges related to data quality
Identify requirements for accurate and consistent data aggregation
Discuss what is needed to prepare the analytic environment to support value based purchasing
6
7
• US 41st in life expectancy
• Japan 1st in life expectancy
• US infant mortality is approximately 4 times Japan
• 28.5 million uninsured in US
*Source: OECD (Organization for Economic Co-operation and Development) 2015
8
The Public View of Value
Source: Health Data Consulting
9
Increasing Measures of Value
Source: Health Data Consulting
*Source:
10
Source: Health Data Consulting
Information Quality
Observations
Accurate
Complete
Consistent Documentation
Coded Data
Well-defined Standards
Accurate implementation
Robust Concept Support
Aggregation
Clear definition
Normalized
Accurate inclusion and exclusions
Analysis
Well defined
Logically valid
Consistently applied
Source: Health Data Consulting
Source: Health Data Consulting Inc.
11
1. Observation of all objective and subjective facts relevant to the patient condition
2. Documentation of all of the key medical concepts relevant to patient care
3. Coding that includes all of the key medical concepts supported by the coding standard and guidelines
Good Patient DataIt’s all about good patient care…
Source: Health Data Consulting
12
Big DataIs more garbage better?
13
Unstructured DataReally?
14
Source: Health Data Consulting
Source: Health Data Consulting
15
Medical ConceptsExpressing the patient condition in codes
Source: Health Data Consulting
Medical documentation scenario:
A [27 year old] [male] patient is seen in [follow-up] for a [Smith’s fracture] on the [right] that was exposed through an [open wound] with [minimal opening and minimal tissue damage]. The fracture has [not healed after 6 months].
Though not explicitly stated in this scenario certain expressions imply other concepts:
“Smith’s fracture” >> [fracture], [radius], [distal], [dorsal angulation], [extra-articular], [displaced]
“minimal opening and minimal tissue damage” >> [Gustilo classification I]
“not healed after 6 months” >> [nonunion]
16
Source: Health Data Consulting
Source: Health Data Consulting Inc.
17
Historical Distribution of ICD-9 Diagnosis Codes3 Years of Data - All claims - All lines of business - 1million Lives
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
5% next5%
… … … … … … … … … … … … … … … … … …
Total Charges by Code3years - $10 Bill
Charge %
Source: Health Data Consulting Inc.
18
Coding SpecificityUnspecified (NOS), Other (NEC) or Symptom/Finding Codes
Source: Health Data Consulting
Code Type Claims Total Charges %Claims %Charges
All Professional Claims 15,352,056 $ 4,030,052,634 100% 100%
‘Unspecified’ (and not ‘Other’ or ‘Symptom
or Finding’)2,902,691 $ 709,765,341 19% 18%
‘Other’ 1,917,163 $ 509,694,935 12% 13%
‘Symptom or Finding’ 3,530,464 $ 675,662,073 23% 17%
Total 'Unspecified', 'Other' and 'Symptom or
Finding'8,350,318 $ 1,895,122,349 54% 47%
Source: Health Data Consulting Inc.
19
Coding SpecificityUnspecified (NOS), Other (NEC) or Symptom/Finding Codes
Source: Health Data Consulting
Code Description Total Charges Claims
78900 Abdominal pain, unspecified site $ 29,331,412 123,737
71946 Pain in joint, lower leg $ 22,973,230 96,786
7295 Pain in limb $ 13,668,722 78,505
78605 Shortness of breath $ 12,533,909 43,463
9597 Knee, leg, ankle, and foot injury $ 9,979,457 41,707
7862 Cough $ 9,250,724 77,430
7851 Palpitations $ 8,181,439 28,228
7820 Disturbance of skin sensation $ 6,531,675 18,238
78060 Fever, unspecified $ 5,269,369 32,603
7823 Edema $ 2,772,549 16,450
Source: Health Data Consulting Inc.
20
Coding PatternsBreast Cancer
Source: Health Data Consulting Inc.
21
Aggregation
The Heart of Policies, Rules Edits and Analytics
Source: Health Data Consulting
22
Source: Health Data Consulting
Information Quality
Observations
Accurate
Complete
Consistent Documentation
Coded Data
Well-defined Standards
Accurate implementation
Robust Concept Support
Aggregation
Clear definition
Normalized
Accurate inclusion and exclusions
Analysis
Well defined
Logically valid
Consistently applied
Source: Health Data Consulting
Source: Health Data Consulting Inc.
23
Aggregating Data - ChallengesSame concept in many places:
Condition Tabular Category Number of Codes
Hypertension Hypertensive Disease 14
Other Categories (14) 115
Pneumonia Influenza and Pneumonia 38
Other Categories (18) 42
Genitourinary Disorders Diseases of the Genitourinary System 587
Other Categories (14) 535
Source: Health Data Consulting
Current categorization in the ICD-10 tabular index
Because of the ‘combination’ nature of ICD-10 codes, they
may not be located in the category the user is expecting
24
Aggregating Data - ChallengesSame concept described many ways:
25
Aggregating Data - ChallengesSame concept described many ways:
26
Aggregating Data - ChallengesSame concept described many ways:
27
Aggregating Data - ChallengesSame concept described many ways:
28
Aggregating Data - ChallengesSame concept described many ways:
29
Source: Health Data Consulting
30
Quality measures
Resource use (cost) measures
Adjustments for risk, severity and complexity– Quality measures
– Outcomes, complication, potentially preventable re-admissions
– Efficiency / utilization measures
Current and evolving payment models dependent on conditions and outcomes of care
Diagnosis impacts?
Source: Health Data Consulting
31
Coding Patterns
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
35.00%
40.00%
% of Claims
42731 Atrial fibrillation
7851 Palpitations
42789 Other specified cardiacdysrhythmias
7850 Tachycardia, unspecified
4279 Cardiac dysrhythmia,unspecified
4270 Paroxysmal supraventriculartachycardia
42769 Other premature beats
42781 Sinoatrial node dysfunction
42732 Atrial flutter
Source: Health Data Consulting Inc.
32
Coding PatternsDysrhythmias
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
% of Claims
N/A Cardiac dysrhythmia, unspecified
42731 Atrial fibrillation
42781 Sinoatrial node dysfunction
42732 Atrial flutter
4260 Atrioventricular block, complete
42761 Supraventricular premature beats
4267 Anomalous atrioventricular excitation
4264 Right bundle branch block
4263 Other left bundle branch block
42611 First degree atrioventricular block
Source: Health Data Consulting
33 3333
Hospital Payment Impacts
Source: Health Data Consulting Inc.
34
Current Distribution of ICD-9 diagnosis codesHistorical Diabetes Coding – 760,556 Claims
Source: Health Data Consulting Inc.
35 3535
Hospital Payment Impacts
Health Data Consulting ©
2014
Source: Health Data Consulting Inc.
36
Concept Based AnalysisDiabetic Retinopathy
Source: Health Data Consulting
Source: Health Data Consulting
Condition Parameter Per person charges* Ratio to Average**
Diabetes $35,341 2.90
Diabetes + Retinopathy $69,424 5.69
Diabetes + Retinopathy + Proliferative $118,654 9.73
* Average total of all claim charges for a person with any claim in this diagnostic category
** Ratio of the average total of all claim charges for a person with any claim in this diagnostic category compared to the average for all persons for all claim charges ($12,200)
Source: Health Data Consulting Inc.
37
Concept Based Analysis
3737
CMS-HCCs
Source: Health Data Consulting
MDMeta © 2016Source: Health Data Consulting Inc.Source: Health Data Consulting Inc.
38
Concept Based Analysis
3838
CMS-HCCs
Source: Health Data Consulting
MDMeta © 2016Source: Health Data Consulting Inc.Source: Health Data Consulting Inc.
39
Source: Health Data Consulting
40
It starts at the source and extends to all stakeholders
Establish the value proposition for data gatherers and coders
– Detail– Standards– Profile comparisons– Connection to clinical care– Empower education with incentives
Focus on accurately representing the precise nature, risk, severity and complexity of the patient condition
Education
Source: Health Data Consulting
41
Define data quality measures and reporting
Data Governance– Empowered from the top– Participation of all stakeholders and contributors
Use data quality measures in a continuous quality improvement process
Tie data quality to value across the organization
Data Quality
Source: Health Data Consulting
42
Assure the right resources are established:– Clinical experts– Financial experts– Coding experts– Data experts– Technical experts– Compliance experts
Clearly define the intended content of all categories of analysis
Aggregation Quality
Source: Health Data Consulting
43
Define the clinical concepts that meet the definition of the category
– What concepts should be included or excluded based on the definition?
Define the code set that:– Includes all codes that should be included– Excludes all codes that should be excluded
QA and share
Ongoing QA, and monitoring in production
Update as standards change and QA requires
Aggregation Quality
Source: Health Data Consulting
44
Healthcare is changing rapidly to an environment that is “value-based”.
The analytic requirements in this new environment will be more focused on the precise nature of the patient condition.
Data quality and aggregation quality are critical to providing reliable, accurate and actionable information to support this new environment.
Summary
Source: Health Data Consulting
45
Getting there will require a data governance
structure and data contributors that are:
– Educated
– Continuously informed
– Incentivized
– Empowered
– Committed
Summary
Source: Health Data Consulting
46
Source: Health Data Consulting
47
Questions
Contact information:• Joseph Nichols MD• [email protected]• 206-478-8227
Thank you !* Please remember to complete your online session evaluations.
48
Speaker Introduction
Thomas HarlanTechnical Team Lead — Reporting ServicesIatric Systems, Inc.
>20 years in healthcare IT
Business Intelligence lead in Epic, MEDITECH, Lawson, and Banner
49
Thomas Harlan
Has no real or apparent conflicts of interest to report.
Conflict of Interest
50
• The BI Mental Model – Efficient Delivery of Actionable Data
– Break around 10:15am
• Best Practices for Data Request Triage
• Best Practices for Report Development
– Lunch at 12:30pm
• Best Practices for ETL Development
– Break around 3pm
• Best Practices for Data Quality Lifecycle
• Summary and Closing remarks
Agenda
51
Recognize the importance of having an enterprise BI architecture
Outline the structure of a BI team and the BI life cycle from data request to data delivery
Define BI best practices
Describe how to minimize the cost of maintaining BI
Learning Objectives
52
An architecture for Delivering Actionable Business Intelligence
Based on standard workflow, standard templates and battle-tested best practices
You are doing some, none, or all of the things we will talk about
Use what you will – or not
All improvement is incremental improvement
Welcome!
53
The Old Approach:
“New request = new printed report”
Reports auto-printed to printers
Reports munged into Excel to feed unknown workflows
Thousands of reports to update
Thousands of reports you’re not sure anyone uses…
= Extraordinary waste
Session I – The BI Mental Model
54
What is the User Story > Action?
Requests for data are expanded to workflow
Requests for data have resulting actions defined
Request triage is conducted
Existing tools are used before implementing new ones
Workflows include data quality (DQ) metrics
We waste as little effort as possible
BI Mental Model – The New Approach
55
Every request is documented via the ticketing system
Focus on the action the end user wishes to take
Do not discuss specific technical tools!
Request is refined by the analyst
Specification of data and source
Can it be done?
Should it be done?
What does the user do next?
BI Mental Model – Request Management
56
BI Mental Model – Request Triage
Follow a standard workflow
Match the intended action to the right tool and system
Communicate status consistently
Hold ticket open until the work is completely done
57
Confirm the data is not already available from the identified
system via a standard output tool
Before building new, see if existing tool(s) can be extended
Avoid duplicating data into parallel systems, if you can
When you have confirmed new is needed, follow best practices for
quick development
Pre-validate the new output before the end-user sees it
BI Mental Model – Reuse First
58
Use the tools you have to the fullest
Follow best practice for each output tool
Establish standard work for development
Set a performance metric
Use a version control system
Pre-plan for updates and upgrades
BI Mental Model – Development
59
During specification build, call out any known data issues
Are they being reported by the DQ process?
During validation, watch for missing data / poor capture workflow
Loop in informatics and add to DQ process
Same goes for end-user validation…
DO NOT fix the data issues in the report code!
Keep your end-users out of Excel, if you can
BI Mental Model – Data Quality
60
Plan and communicate an expectation that all BI staff will, in time…
Follow standard work for all development
Be able to use all available tools
Be able to address any request
Be able to meet the performance metric
This means a comprehensive training and role rotation plan.
BI Mental Model – Training Staff
61
How long does it take to be comfortable with a system?
Define your process of turning User Story > Action
Reinforce the use of standard work
Define what tools are in use
Implement training and role rotation
Mentoring
Implement a technical career path
Burgers must be flipped, but there must also be new cheese
BI Mental Model – Retaining Staff
62
BI Mental Model – Discussion
63
Be back in 15 minutes
Break!
64
Best Practice – Data Request Management
Starting from a ticket, we:
Define the User Story
Refine the data specification
Triage the request to available tools
Develop (if necessary):
- New coding
- Valid cycle
Closeout
65
Define the User Story
Who is the user, and what workflow are they trying to improve?
What data do they need to act upon?
How does that data need to be presented?
Which system should it come from?
What actions are they going to take, based on that data?
What are the next steps, beyond initial actions?
How do we identify data quality issues?
Best Practice – Data Request Management
66
Refine the Data Specification
Even reports have a data specification… but it is essential for extracts
Use a standard template to document the spec
Let the template do some of your work for you
Encourage the requestor to define the specification
Store the filled-in template with the code, in version control
Send the specification to development, with the ticket or task
Best Practice – Data Request Management
67
Triage the request to available tools
Once the specification is set (which may be quite simple)…
Is there a tool (report or extract or dashboard or…) which will provide
the user with the data they need to act?
If there is, note this in the ticket and re-route to the user.
If not…
Use the appropriate system triage map to determine what tool to
develop in (see next slide):
Best Practice – Data Request Management
68
Best Practice – Data Request Management
69
Develop (if necessary): New Coding
Assign a Data Request Number (DRN) to the new development
Define or modify a version control project to include your new code
Leverage existing code where possible – infrastructure objects!
Build in a consistent way, using the best practice templates
Update your ticket or task to reference the setup
For extracts, a wiki or knowledge-base article is critical
Best Practice – Data Request Management
70
Develop (if necessary): Validation Cycle
Never send data via a tool to a user, unless it has been pre-validated
If an analyst has done the specification prep, they hand off to the developer
The developer works directly with the end-user on:
Validating the data
Identifying data quality scenarios
Best Practice – Data Request Management
71
Best Practice – Data Request Management
72
Closeout
Check code into version control
Publish the report, or schedule the extract
Update the Data Request Number in tracking
Update the ticket or task
Have the beverage of your choice!
All of this is standard work for your BI team. Checklist it, and
do it every time.
Best Practice – Data Request Management
73
Best Practice – Data Request Management
Discussion
74
Be back in 15 minutes
Break!
75
Foundation Concepts
Correct data … Fast to run
Correct data … Well presented
Correct data … Easy to maintain
Best Practice – Report Development
76
Reporting Architecture
Drive reports from stored procedures
Deploy reports via the web
Work to a performance metric
Parameterize with data-driven lists
Establish and match a site style guide
Best Practice – Report Development
77
Component Naming
Reports will be composed of at least two components (query and layout)
Use your Data Request Number (DRN) in the name of each component
Include the DRN on the report layout as well
Remember you have Infrastructure Numbers for common objects
Best Practice – Report Development
78
Best Practice – Report Development
Server Environment
DATASERVER
Production data repository database
RPTSERVER
Web server providing viewable / scheduled reports
FILESERVER
Offline copy of all report components
79
Report Deployment
On the RPTSERVER, take care to arrange your reports into a clean folder structure and…
Only assign access permissions by groups at the folder level
Report server groups should link to AD groups
Where possible, use short-cuts or linked reports from a hidden
“main” folder
Best Practice – Report Development
80
When picking the right reporting tool, we need to consider:
What latency is suitable for our end-user using the report?
Where does the report need to be visible?
Is the data reportable?
Do we need to click-through to the chart or account?
Does the output need to be scheduled?
Do we have to go back quarters or years for data?
Report Development – Tool Selection
81
Report Development – Tool Selection
82
Developing in TEST is almost always a waste of time
…Unless the build you’re reporting against is only in TEST
Use a DEV instance instead
DEV has many advantages
Report Development – Live vs. Test
83
1. Create (or re-use) a stored procedure, from a standard template
2. Stored procedure is always parameterized
3. Parameters are driven by re-usable stored procedures
4. Code header is in standard format
5. Output is validated before report layout build
6. For a new report layout, start from a standard template
Report Development – Step by Step
84
7. Format the report attractively
8. Use dynamic features wisely
9. Publish report to \TEST folder for end-user validation
10.Once validated, move to \LIVE folder for general use
Report Development – Step by Step
85
Lookup lists should be driven by stored procedures as well.
Do not embed code in the reports
Use an Infrastructure DRN number
Maintain them in version control
When an update comes, you want to easily identify and manage
changes to lists used by hundreds or thousands of reports.
Report Development – Lookup Lists
86
In every stored procedure you build, be mindful of the future.
Start from a template
Parameterize for facility, start date, end date, department(s) and so on
Template code can process parameters
Comment out what you don’t need
Report Development – Core Query Setup
87
In every report layout you design, be mindful of the future.
Start from a template
Include a URL driven logo
Parameters are data-driven and provided by the template
Standard header, footer, fonts and colors – style sheet!
Report Development – Common Layout
88
Meet your performance metric!
Don’t use DISTINCT in the report
Avoid sub-reports processed per row
Don’t filter data in the report
Split the data-processing into the stored procedure, save the report
layout for formatting and interactive layout.
Report Development – Common Problems
89
Use an enterprise-wide style sheet for data output
Output must be attractively presented, or it may be wasted
Assume your report will be printed, and presented to the Board
Report Development – Layout Appearance
90
Codes are lovely, but most of the time they’re not easy to understand
Name [ Mnemonic ]
Report Development – Mnemonic Display
91
Identifying a patient properly is key to almost every clinical workflow:
This is not enough:
Doe, John [ 00456 ]
This is best:
Name [ MRN ] (Gender/Age) AccountNumber
Report Development – Positive Patient ID
92
Report Development – Discussion
93
LUNCH!
Be back at 1:30PM
94
SQL Server Integration Services
Correct Data… Automated
Correct Data… Monitored
Correct Data… Reported
Follow the BI Mental Model to define the User Story data specification
before coding!
Best Practice ETL – Overview
95
1. Automate and schedule entire process
2. Provide run-on-demand
3. Set alerts for failure
4. Data Flow tasks driven by stored procedures
5. New db objects stored in a custom catalog
6. Stored procedures are parameterized
7. EXEC command generated by a control variable
Best Practice ETL – Common Components
96
8. Package sends files securely
9. Tab-delimited files are preferred
10.Meet the performance metric!
11.Notify the customer of work done
Best Practice ETL – Common Components
97
Three Servers?
Data Repository database server
SQL Agent with Integration Services
Off-line copy of extract code
Three File Storage Locations:
Production location of scheduled jobs
Offline copy of projects and documentation
Output files to send, and archive of sent files
Best Practice ETL – SSIS Environment
98
Track ETL requests with a Data Request Number.
Use the DRN as a common identifier:
C0200-PressGaney.sln
C0200-PressGaney.dtsx
C0200-PressGaney-SP.sql
Best Practice ETL – Data Request Tracking
99
Custom Database objects go into:
zcus
Work against:
DEV, if you have it, or LIVE otherwise
Best Practice ETL – LIVE vs. TEST vs. DEV
100
Stored procedure per file format
SP result-set matches column-for-column to the
specification
SP results match file specification for type and
length per column
SP’s collect by date range and facility id
Best Practice ETL – Source Data
101
Best Practice ETL – Discussion
102
One DTSX package
Control variables
Data Flow task per file
End to end automation
Best Practice ETL – Extract Components
103
One DTSX per process
Control variables
Load to staging
Delete matching keys
Insert from staging
Best Practice ETL – Datamart Components
104
One DTSX per process
Global control variables
Load to Staging
Delete Matching Keys
Insert to reporting tables
Best Practice ETL – Loader Components
105
A Bad Scene…
106
107
Central shared storage of projects:
\\FILESERVER\ETL\Live
Using DRN-prefixed folders:
\\FILESERVER\ETL\Live\C0200-Press-Ganey
Best Practice ETL – Managing Projects
108
Project Standard Structure:
– \Documentation
– \SSIS
– \SQL
– \Utilities
Match to version control project structure.
Best Practice ETL – Managing Projects
109
When creating a new project:
Use a short folder path, with short file names
EncryptSensitiveWithPassword
Don’t work on your drive C:!
Use a template DTSX file
Best Practice ETL – Project Creation
110
Before the first Data Flow task:
Create control variables for the extract path, start/end
dates, facility id(s), database server name, etc.
Create two or three variables per file being generated:
Filename, Query, Rowcount
SSIS 2012 + Package variables can be overridden by
package parameters
Best Practice ETL – Package Control
111
For each database connection:
Set Packet Size
Set Commit Size
Set Batch Size
Best Practice ETL – Configuring Connection(s)
112
For each task:
A meaningful name
Has an OLE/DB source
Captures a row count
Has a destination
Use the control variables!
Best Practice ETL – Working Tasks
113
Generate all file names:
Start from an @cExtractPath variable
Create custom @cFileName(s)
Override the ConnectionString property
Best Practice ETL – File Management
114
Start SSIS stored procedures with:
SET NOCOUNT ON ;
SET ANSI_WARNINGS OFF ;
Add to your template!
Best Practice ETL – SQL Setup
115
#temp tables in stored procedures called by SSIS packages generate an error like:
"Invalid object name '##Payment'.".
Best Practice ETL – Temp Tables
116
You fix this one of two ways:
Pre-SSIS 2012, use a result-set “template” in your code
Post-SSIS 2012, use WITH RESULT SETS clause
Best Practice ETL – Temp Tables
117
Best Practice ETL – Discussion
118
Users like job notification emails:
File names created, with rowcounts
Folder or destination the file went to
Time spent processing the extract
Best Practice ETL – Job Success Emails
119
On all SQL Agent jobs, configure failure notifications:
Email alert when the job fails
What was the error?
Best Practice ETL – Job Failure Emails
120
In a production environment, there are a lot of job emails!
Use your email rules to sort and re-prioritize the messages:
SUCCESS goes to an \Archive folder
FAILURE promote to High Priority and flag in your inbox
Best Practice ETL – Managing Job Emails
121
With many extracts running each day:
Have them run without contention
Be self-documented
Be easy to manage
Configure job streams, with packages running in sequence.
Best Practice ETL – Scheduling
122
Log job run history to custom table
Use this history table to:
Trigger alerts
Best Practice ETL – Job History Tracking
123
Here is a daily job stream w/ four scheduled jobs:
Schedule Name Step Package Run Time
Daily – 0700 - Multistep 1 M0251-Studer Group Export.dtsx At 0700
2 M0999-Safety Surveillor RX.dtsx
3 M0001-15 Minute Census Snapshot.dtsx
4 M300-EmCare Billing Analysis.dtsx
Best Practice ETL – Scheduling Example
124
SQL Agent Jobs are only listed by name. Make that name useful:
(Recurrence) - (Time Pattern) - (Package Name or "Multistep")
Best Practice ETL – Schedule Names
125
When scheduling jobs in SQL Agent:
Use an SSIS Proxy Account
Use Windows authentication
Run in 64-bit mode
Reuse existing schedules
Best Practice ETL – Schedule Setup
126
Listing of all ETL jobs
With links to documentation
With dependency information
The Job Book lets us recreate the scheduled job
environment if we lose the SSIS server.
Best Practice ETL – The Job Book
127
Is your reporting database recovery model set to
SIMPLE?
If set to FULL, and you are doing large extracts, this
may cause disk space issues.
If this occurs, change recovery to SIMPLE or BULK
LOGGED for the duration of the process.
Best Practice ETL – Transaction Log Growth
128
Data Request Number entry
Extract Wiki/KB Article
FILESERVER ETL Folder
Use SCHEDSERVER Catalog or Pkg Folder
Use SCHEDSERVER Extract and Archive Folder(s)
Update your Job Book
Update the CMDB CI (if you have a CMDB)
Best Practice ETL – Closeout
129
Best Practice ETL – Discussion
130
Be back in 15 Minutes
BREAK!
131
Work queues
Standard reports
Custom reports
Can you take action directly from the report?
Can you measure DQ progress and attribution?
Best Practice Data Quality - Overview
132
Incorrect data exists in the system
It goes out to a vendor via an ETL file
A vendor error report comes back
Your staff fix the bad data in the vendor system
And… repeat the next month
Best Practice DQ – Common Workflow
133
The Patient Record
Test patients
Visits without assigned facility or service area
Patients who have expired on the visit, but not on
the patient
No MRN
ADT
Too many census records
Best Practice Data Quality – Common Issues
134
Clinic Workflow
Checking out patients
Hospital Billing
CPT Codes from external LAB interface
Recurring visits not attached to recur. accounts
LOS via bed charges vs. admit-discharge days
DRG not populating from grouper interface
Best Practice Data Quality – Common Issues
135
Hospital Billing, Continued.
State-specific DRG and CPT Codes
Placeholder provider IDs
Account discharged before admitted?
Accounts discharged in the future
Accounts discharged without Discharge Disposition
Best Practice Data Quality – Common Issues
136
Providers
Provider group data management (membership, phone, address)
Provider records with shared/duplicated DEA or NPI Numbers
Data from Lab System
Collector User ID instead of freetext name
Discrete or component results
Best Practice Data Quality – Common Issues
137
On some systems …
Reports can click-through to the account / visit if
properly configured
Which is fantastic… but:
No Metrics about DQ errors and resolutions
We don’t know who did what, when or if it
recurred
Best Practice Data Quality – What to do?
138
We want an architecture:
Runs daily
Captures metrics of DQ events
Generates action reports
Batches to data owners via email
Provides area-specific dashboard
End-user Reports include a “bad data” link
Best Practice Data Quality – What to do?
139
Identify data owners
Categorize sources of truth as you find them
Identify their top 1 issue
Define DQ workflow, using 1 issue
Deploy DQ report, metrics, etc. for 1 issue
Wait for tickets requesting new DQ items
Best Practice Data Quality – What to do?
140
Best Practice Data Quality – Discussion
141
Contact Information
Thomas Harlan
1-978-674-8330 (PST)
Remember to complete your online session evaluations!
Questions:
Please use blank slide if more space is required for charts, graphs, etc.
To remove background graphics, right click on selected slide,
choose “Format Background” and check “Hide background graphics”.
Remember to delete this slide, if not needed.