flexner report (2010) data analytics in the clinical · data analytics in the clinical laboratory...

9/5/2016

1

DATA ANALYTICS IN THE CLINICAL LABORATORY: ENHANCING DIAGNOSTIC VALUE AND REDUCING COSTS

Brian Jackson, MD

Jason Baron, MD

Anand Dighe, MD, PhD

DISCLOSURES

• Brian Jackson– ARUP Laboratories (Nonprofit entity of University

of Utah): Consultant/Salary

• Jason Baron– Research support to his instituion (MGH) from

IBM

• Anand Dighe– No relevant disclosures

Data Analytics in the Clinical Laboratory Part 1:Measurement Philosophy

Brian Jackson, MD, MS

VP, CMIO, ARUP Laboratories

Assoc Prof of Pathology (Clinical), University of Utah

Flexner Report (2010)

• Established science as foundation of medical training

• Created cultural divide between physicians (science) and administrators (business)

9/5/2016

2

Myths: Business versus Medicine

• Myth #1: Business management is primarily about money

– Reality: Business management is primarily about organizational effectiveness

• Myth #2: Mixing medicine and business is “dirty”

– Reality: If we really care about our patients, we’ll use all useful tools to improve the effectiveness of our healthcare organizations

Is your laboratory successful?

Prove it.

Ideal laboratory metrics:

• Support learning and improvement

• Summarize major dimensions of

performance

• Be balanced (unbiased)

• Include both outcomes (past performance)

and processes (future performance)

3 Key Questions

• How much are we accomplishing?

• What does it cost us to accomplish it?

• How reliable are our processes?

9/5/2016

3

3 Key Questions

• How much are we accomplishing?

– Most businesses: Revenue

– Value-based healthcare: Patient benefit

• What does it cost us to accomplish it?

– Resources ($)

• How reliable are our processes?

– Quality

Organizational Learning and Improvement

• Senior leadership needs:

– Big picture

– Representative

– Don’t get lost in the details

• Front line needs:

– Directly tied to day-to-day activities

– Under their control

Putting it all together (Factory)

Quality Revenue Costs

Key rollupmeasure(s)

Sales Total costs

Defects by category

Market shareRevenue by segment

LaborSuppliesDepreciation

Defects by process

Revenue per sales rep

Detailedcosts per department

Executive

Front line

Putting it all together (Clinical Laboratory)

Quality Patient Benefit Costs

Overall health system reliability

Global benefit Total cost to lab

Per clinical practice unit

Benefit per test Cost per test

Per test:• TAT• Accuracy• Process

quality

Benefit per case• Variation• Consistency

with guidelines• Consistency

w/expert opinion

Cost per case

Executive

Front line

9/5/2016

4

Where Do We Have Good Metrics Today?

Executive

Front line







quality



w/expert opinion

Cost per case

Where Are the Opportunities?

Executive

Front line







quality



w/expert opinion

Cost per case







quality



w/expert opinion

Cost per case

Managing Diagnostic Test Utilization

Executive

Front line

Total Cost of Laboratory Operations

• Labor

• Reagents

• Instruments

• Facility overhead

– Space, utilities, IT, etc.

9/5/2016

5

Cost per Test

• Proper Approach

– Labor, reagents, instruments, overhead

• Do not use 3rd party fee schedule!

• Do not use chargemaster!

Cost per Case

• Assumes you have valid costs at component level

• Overhead allocation is tricky

• Dependent on the clinical algorithms

Harvard Business Review Sept 2011


Executive

Front line







quality



w/expert opinion

Cost per case

9/5/2016

6

Global Measures of Healthcare Quality?

Program # Measures # Diagnostic # Lab

HEDIS 74 20 9

CMS ACO 33 13 4

Choosing Wisely

135 90 21

Patient Benefit per Test

• Function of how the test is used

– NOT an intrinsic quality of the test itself

• Example: H pylori testing

– Do stool Ag and breath test provide more patient value than serology?

– Answer: Depends on the rate of endoscopy

Holmes et al. BDM Health Services Research 2010, 10:344

Patient Benefit (of a Test) per Case

• Outcomes

– Generally not practical in this setting.

• Normative (Evidence Based Medicine)

– Guidelines

– Other clinical literature

– Local expert opinion

• Non-normative/Descriptive

– Variation


Executive

Front line







quality



w/expert opinion

Cost per case

9/5/2016

7

Data Analytics in the Clinical Laboratory Part 2

Using Analytics to Guide Operational and

Clinical Decisions

Jason Baron, MD

Medical Director, Core Laboratory

Massachusetts General Hospital

Assistant Professor

Harvard Medical School

Overview

Lab Informatics in Quality Management

1. Using metrics to guide clinical and operational decisions

• Ask useful questions

• Maintain usable data

• Address the question and appreciate limitations

• Variation Analysis

• Take a panoramic view

2. Computational Pathology and emerging

• Example: Spurious glucose identification

• Example: Computational Pathology and ferritin prediction

Overview











Metrics should be designed around a

specific question

• Effective metrics are designed purposefully and thoughtfully–

to answer a specific question

• Excessive use of metrics not designed around key questions

may just lead to data overload without real added value

9/5/2016

8

Areas to ask questions

Domains

• Operational

• Analytic and pre-analytic performance

• Financial performance

• Clinical performance

• Utilization management

• Computational pathology

Circumstances

• Before vs. after an intervention

• Trends over time

• Comparisons to benchmarks

• Goal-based

• Standards-based

Operational Metrics Examples

Examples

• Turn around time

• Tech productivity

• Results per hour

• Can be broken down in many ways (shift, section, etc.)

Collect to result time (days) for a test

Analytic and Pre-analytic Performance

Examples

Examples

• Result distribution

• Frequency of anomalous results reported

• Corrected results

• Frequency of outlier results

Plasma Potassium Example

Example : Babesia Serologies by Physician

0

20

40

60

80

100

120

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z a b c d e f g h

Test

s

Provider

Positive

Negative

9/5/2016

9

Example: Goal and Standards based

Metrics

March April

ER 83 84

IP 87 87

OP 90 89

Percent of Specimens

with a Valid Collect Time

• Can be metrics that have a

right answer—or best

performance

• E.g. no specimens received

without a valid collect time or

no corrected reports is optimal

• Can set realistic “red, yellow,

green” or other cutoffs to

define realistic goals

Overview











Data: The Lifeblood of Analytics and Metrics

Raw

DataMetrics

Question Information

Proper Interpretation

•Metrics and analytics require a source of data

•Obtaining this data is often a frustration for laboratories

•Laboratories should advocate for direct data access

Key: Data Accessibility

Text

File

LIS

Nightly

reports

Datamart

Digest and

Import

Pathologist

ODBCSQL

Queries

•Data readily available

•Can easily link data from multiples sources for analysis

EHR

Nightly

reports

9/5/2016

10

Overview











Metrics by themselves are just numbers

• Need to ensure that the metric truly addresses the intended question/hypothesis/problem

• Almost all metrics are subject to limitations

• Standard must not be perfection

• Remember that not everything is measureable

• Be aware of key limitations

Confounding and Effect Misattribution

• Correlation ≠ causation

• Frequently, metrics are designed to assess the impact of an initiative

or track performance over time

• Generally multiple factors vary before/after an initiative or over time

• Changes in a metric may be misattributed to the wrong cause

• Normalization

– E.g. tests per visit or tests per admission

• Adjust for seasonal variation

• Leverage statistics

• Sometimes it is sufficient to be aware of and accept limitations

Strategies to Controlling Confounding

9/5/2016

11

Seasonality Example

•Initiative in the summer of 2014 to restrict babesia serologies

•Although maybe these data speak for themselves, capturing the full effect

requires considering seasonality

Overview











Variation Analysis: An Important Time to Consider

Confounding• Test ordering patterns can be compared between physicians and leveraged as a

utilization management tool

• Can be used as a tool both to identify utilization improvement opportunities and to encourage thoughtful utilization

• Clinicians identified as outliers may adjust test ordering practice to better mimic colleagues

• Important to qualitatively or quantitatively account for factors that appropriately impact utilization (e.g. subspecialty or patient mix)

Inter-Specialist Variation in Sendout Costs

9/5/2016

12

Sendout Adjusted Expenditure, Accounting for Diagnosis Other Variation Metrics

Yield Analysis

•Variation analyses can look at metrics besides cost

•Examples

•Counts

•Yields

•Appropriateness

•Outcomes

Remember

• Not all outcomes can be measured

• Some metrics are qualitative

Overview











9/5/2016

13

Ensure Metrics are Balanced

• Management often entails balancing competing priorities

• It is important to track competing goals within the lab in parallel

• Otherwise management initiatives may too strongly favor one goal at the expense of others

Cost

Reduction

Quality

Improvement

NB: Cost and quality are not necessarily

in competition but they can be

Potential examples of balancing

Cost Quality/ Service

Full time Staff Overtime expense

Utilization reductions Clinical outcomes

Capital expenses Operating expenses

Data accessibility Data security

Important to consider downstream effects

Overview











Raw

Most Processed

“Given this patient’s test results and clinical data,

administration of vancomycin will improve odds of

survival from 37% to 87%.”

“EBV-positive immunoblastic reaction,

consistent with infectious mononucleosis”

Present

Comp

Path

Atomic data

Interpretive

comments

Diagnoses

Integrated

Information

Predictive

Information

103127

3.8

101

21 1.86

19

9/5/2016

14

Overview











Computational Derivation of Knowledge: Example

Spurious Glucose Identification

Spuriously

elevated glucose

result

•Commonly problem at many hospitals

•We were seeing spurious critically elevated glucose

results about once per day

•Fewer than 10% of these spuriously elevated

glucoses were being identified

Goal: Develop an Algorithmic Protocol to Distinguish

Spurious from Real Critically Elevated Glucose Values

Spurious Glucose Identification, Methods

Annotated Training Data (glucose >500 mg/dl)

Patient Glucose Na Additional Predictors

(K, CO2, AG, etc.)

Gold Standard Annotation

A 670 119 … Spurious

B 710 141 … Real

C 721 138 … Real

… … … … …

Supervised

Machine Learning

(Recursive

partitioning

decision trees)

Decision Tree

Test data or Un-annotated patient data

Patient Predictors

(glucose, Na, K, AG, etc.)

X …

Prediction as to whether

result is real or spurious

AJCP 2010 133:860

Tree built Using: Na, K, Cl, Bicarbonate, Anion Gap,

Glucose, and 30 day mean glucose

Spurious Glucose Identification, Results

Training Data

Test Data

Spurious

Correctly

Classified57 32

Total

Spurious61 37

Sensitivity

(95% CI)93%

(84-98%)86%

(72-95%)

Real

Correctly

Classified68 5

Total Real 77 6

Specificity

(95% CI)88%

(79-9%)

83%

(42-99%)

AJCP 2010 133:860

9/5/2016

15

Overview











Example: Ferritin Prediction

We use Ferritin as an analyte of focus in an early proof-of-

concept

•Ferritin

•A marker of iron stores

•Used in the diagnosis of iron deficiency

•must be interpreted in the setting of other clinical and

laboratory data

•thus a good problem for data integration

•Decreased in iron deficiency

•Increased in inflammation

Ferritin Methods OverviewRaw Data (3 months outpatient

ferritin values)

•Transform ferritin values

•Divide at random into training and

test partitions (7:3 ratio)

•Mask ferritin results for test partition

Raw Data with

ferritin masked for

test patients

•Impute missing data

•Use 4 different imputation

methods

“Completed”

dataset

Performance

Metrics

•Predict ferritin results

•4 regression methods

•1 classification method

•Pair each with each

imputation method

Predicted ferritin

results and

classifications

•Compare predicted ferritin

results to measured results

•Compare predicted to

“masked” results for test

partition

•Review selected cases to

determine clinical

significance

AJCP (2016). 145:778-88.

Ferritin Classification PerformanceTest Data

Negative Control

AJCP (2016). 145:778-88.

9/5/2016

16

Ferritin Case Review

Test dataset (N=1538)

Ferritin

Results

Predicted

Ferritin

Results

Ferritin result differs

from prediction by a

factor of 10 or more

Highly discrepant

results

N=26 (1.7%)

4 Cases

Predicted ferritin

≤30 ng/ml (male ref

limit)

Case Ferritin Predicted

Ferritin

Impression Comment

1 230 21 Iron deficiency,

not clinically

identified

Ferritin increased

secondary to inflammation

2 197 19 Recovering iron

deficiency

Receiving IV iron therapy

3 1768 9 Limited predictive

data

•Only two predictor tests

available

•Decision support will

likely require a minimum

number of predictor tests

4 197 19 Complex

hematologic

picture

Referral to hematology

would have likely been

useful had the testing been

ordered by a non-specialist

Conclusion: Predicted ferritin may more accurately reflect

underlying iron status in some patients� signals potential

application to clinical decision support

AJCP (2016). 145:778-88.

Ferritin Summary

•Coexisting data can discriminate normal from abnormal ferritin

results with a high degree of accuracy (AUCs as high as 0.97)

•Predictions of numerical ferritin results were moderately

accurate

•In at least certain cases, predicted ferritin may better represent a

patient’s underlying iron deficiency status.

Final Conclusions

1. Analysis of existing data can identify quality improvement opportunities

and strategies

2. Metrics must be developed thoughtfully and strategically

3. Artificial intelligence, machine learning and big-data analytics provide

emerging opportunities to enhance diagnostic precision and clinical quality

Acknowledgements

• Anand Dighe

• John Gilbertson

Ferritin Prediction

• Peter Szolovits

• Yuan Luo

• MGH eCore– MGH-

MIT Grand Challenge

Grant

• Kent Lewandrowski

• Joseph Rudolf

Many Aspects

Spurious Glucose Identification

• Craig Mermel

Variation Analysis

• Jeffrey Weilburg

• Michael Hidrue

Babesia analysis

• Vikram Pattanyak

9/5/2016

17

Data Analytics in the Clinical Laboratory Part 3

Using Data Analytics and the Electronic Heath Record to Enhance Quality and Safety

Anand Dighe MD, PhD

Director, MGH Core LaboratoryDirector, Laboratory and Molecular Medicine Informatics

Associate Professor, Harvard Medical School Massachusetts General Hospital

Boston, MA

Inter-pretation

ReportingProcessing/AnalysisCollectionOrdering

Post-AnalyticAnalyticPre-analytic

Middleware

• Interference checking

• Rules-based auto-dilution

• Automated add-ons

Test Result Auto-verification

Info Buttons

• Guidelines

• Literature

• Online resources

PathologyInterpretative

Services

PROCESS

Computerized Provider Order Entry (CPOE)

• Test panels

• Redundancy alerts

• Clinical guidelines

Automated Specimen Collection Process

RFID/bar coding

• Enhanced Electronic Medical Record systems

• Actionable result reporting

Institutional Reflex

Algorithms

Enhanced Result

Generation

Informatics in the Laboratory Testing Process

Why Emphasize Informatics?

• Acts as a force multiplier for many projects– Orders. Growth of the EHR and CPOE provides the

opportunity for informatics approaches to improve order entry

– Results. Enables high levels of robotics and auto-validation to be safely and efficiently implemented

– Results Management. Informatics approaches to results management can provide added quality and safety

• The increased growth of informatics-based approaches requires new strategies to measure lab and EHR outcomes

2010: in the U.S. only 12% of hospitals utilize computerized provider order entry for laboratory testing

2016: “Meaningful use” initiatives and other health care reforms have dramatically changed this to an estimated 50-70% usage in just 5 years

Advantages of CPOE• Opportunity to interact with the ordering clinician in

real time• Can present information to the clinician at the time of

the decision• Education is much less effective before or after the decision

• Links ordering physician tightly with the order, simplifying utilization audits

Computerized Provider Order Entry (CPOE)

9/5/2016

18

Inpatient/ED/Outpatient Provider Order Entry

Providers

CPOE: Pathology’s Perspective

Laboratory Information System

Laboratory Staff

Lab Orders

Pathology must avoid from being “shut out” from the CPOE system

• Improved diagnosis

• Error reduction

• Utilization control

• Laboratory efficiency

Providers

MGH PathConnect Middleware

Laboratory Staff

Permits Pathology to have control over Provider Order Entry screens



Lab Orders


Web

services

MGH PathConnect MiddlewareKnowledge Management

• Synchronizes with Laboratory Information System (LIS)

• Receives data from LIS regarding each test

• LIS test data can be augmented with ordering messages, alerts, search terms, related tests

• Allows cataloging of Pathology data such that it can be shared with other parts of the organization via web services

POE LISMGH PathConnect

• Provider order entry calls middleware web service to build test dictionaries in a “just in time” manner

• Screen content can be updated in real time by Pathology via a web service

• Critical to control the content of menus to control utilization

POE Lab Ordering ScreenThe order entry group leaves us “white space” that the lab fills in.

User Interface Built Entirely from MGH PathConnect Data

9/5/2016

19

Order Entry: the Importance of Search

• Support synonyms

• Support misspellings

• Provide key test information (TAT, cost)

With 1,500 tests on the menu a robust search engine must be a part of all test order entry applications.

Provide information to guide appropriate utilization

Improving Vitamin D Utilization with CPOE

Search provides more than a list of possible matches �

Provides information to guide appropriate utilization

Middleware enables rapid (minutes, to author and update test) responses to utilization issues

• Adding non-interruptive ordering message dropped 1,25 OH vitamin D orders by 70% (p < 0.001)

• Cost savings = $20K/yr

MGH PathConnect Middleware(5 minutes to create and test new alert with no IS involvement)

MGH Order Entry Screen

Passive, non-interruptive ordering message

9/5/2016

20

CK-MB Additional Info Screen (Interruptive Alert)

Inpatient CK-MB Results Per Day

0

20

40

60

80

100

120

140

1/18

/201

1

1/25

/201

1

2/1/20

11

2/8/20

11

2/15

/201

1

2/22

/201

1

3/1/

2011

3/8/

2011

3/15

/201

1

3/22

/201

1

3/29

/201

1

4/5/

2011

4/12

/201

1

4/19

/201

1

4/26

/201

1

• Sustained 80% reduction in CK-MB orders within 3 weeks

• Cost savings of $30,000 per year

Added interruptive alert to POE

Building systems that get smarter

1) A strong vocabulary and data model can make getting to version two unnecessary

2) Monitoring actual use is critical to understanding how to improve the underlying data model

– Create monitoring reports for use on Day 1

– Expect and embrace failure!

– If you have a system that can learn you need content

1

Provider Order Entry

Cache

Lab Orders

Providers

Sunquest LIS

Laboratory Staff


ODBCXMLMS SQL

Web

services

Daily reports of all searches and orders

Expect Failure: Monitoring Reports for Orders and Search

2a) Non-productive searches

b) Free text orders

9/5/2016

21

Analysis of User Search Productivity

Daily reports of all user searches, results, and orders available to lab

Reasons for Nonproductive Searches

FIX: Update middleware with misspellings

FIX: Update middleware with synonyms

FIX: Add test via middleware

(e.g. “insulin”)

(e.g. “celiac”)

(e.g. “syfillis”, “ferritan”)

1


Cache

Lab Orders

Providers

Sunquest LIS

Laboratory Staff


ODBCXMLMS SQL

Web

services

Daily reports of all searches and orders

Expect Failure: Monitoring Reports for Orders and Search

2a) Non-productive searches

b) Free text orders

Free Text Orders (e.g. “workarounds”)

0.0%

2.0%

4.0%

6.0%

8.0%

10.0%

12.0%

14.0%

6/2

5/2

008

7/2

5/2

008

8/2

5/2

008

9/2

5/2

008

10/2

5/2

008

11/2

5/2

008

12/2

5/2

008

1/2

5/2

009

2/2

5/2

009

3/2

5/2

009

4/2

5/2

009

5/2

5/2

009

6/2

5/2

009

7/2

5/2

009

8/2

5/2

009

9/2

5/2

009

• Free text (directly typed) orders monitored on a daily basis by Pathology

• Free text is essential to eliminate before orders are communicated electronically

• Permits near real time intervention with non-compliant physicians and/or system changes (adding tests, improving search)

• Current free text percentage 0.32% (4 sigma)

Order communication start

9/5/2016

22


• Provider Order Entry is a key leverage point for Pathology to improve ordering practices, prevent ordering errors, and avoid pre-analytic error

• Pathology should have input/control over all laboratory order entry modules (IP/ED/OP) to permit rapid responses to ordering issues

• Centralization of laboratory knowledge combined with an understanding of ordering behavior is essential to improve quality and control utilization

Clinicians

OrderingClinicalInterpretation

Collection

Receiving

Resulting

InterpretiveReporting

Processing

Lack of robust electronic order entrysystems for all hospitals and all sites of care (inpatient, outpatient, ED) and for all pathology areas (AP, core, micro, blood bank)– Resulting manual processes

create huge amount of rework and inefficiency

– Hampers innovation and customer service by consuming staff resources

– Prevents implementation of decision support tools

Lack of closed loop results acknowledgment. – From ordering to resulting to

action taken need to ensure lab results are having their desired outcome

Order Entry and the Enterprise EHR


Providers

CPOE: Pathology’s Perspective


Laboratory Staff

Lab Orders

Pathology must avoid from being “shut out” from the CPOE system

Clinical Service BWH FH DFCI MGH1 NSMC NWH

Clinical Laboratories-Chemistry-Hematology-Microbiology

BICS2 MEDITECH

v5.64

Sunquest

v6.3; upgrading to v7.1 in 2013

Sunquest

7.1

Sunquest

v6.2; upgrading to v7.0 in 2013

MEDITECH

v5.64

Phlebotomy Collection Lattice MEDITECH

v5.64

Sunquest

Collection Manager

Sunquest Collection Manager

Will be Meditech

Point-of-Care Testing Abbott/Sybase

ISTAT

Roche Does Not Perform

Telcor Abbott/Sybase3 RALS

Anatomic Pathology-Surgical Pathology-Cytology-Cytogenetics-Molecular Diagnostics-Autopsy

Sunquest

- Powerpath

Cerner/Co-Path

v3.2

N/A – send specimens to

BWH

Sunquest

- Co-Path

Sunquest

CoPath

V4; upgrading to V6 in 2013

MEDITECH

LABLION Specimen Tracking

Blood Bank &

Transfusion Medicine-Donor Center-Blood Bank

Mediware

-Lifetrak

-HCLL

MEDITECH

v5.64

Sunquest

v6.3 for Cell Therapies only (limited use).

Use BB at BWH

Mediware

-Lifetrak

-HCLL

Sunquest MEDITECH

v5.64

Tissue Typing G4 N/A N/A – send specimens to

BWH

mTilda N/A5 N/A5

Reference Lab Mayo7

LIS Systems Across PHS (2013)

9/5/2016

23

Clinical Service BWH FH DFCI MGH1 NSMC NWH

Clinical Laboratories-Chemistry-Hematology-Microbiology

Sunquest

7.1

Sunquest

7.1

Sunquest

7.1

Sunquest

7.1

Sunquest

7.1 in Jan 2017

Sunquest

7.1

Phlebotomy Collection Sunquest

Collection Manager

Sunquest

Collection Manager

Sunquest

Collection Manager

Sunquest

Collection Manager



Point-of-Care Testing Abbott/Sybase

ISTAT

Roche N/A Telcor Abbott/Sybase RALS

Anatomic Pathology-Surgical Pathology-Cytology-Cytogenetics-Molecular Diagnostics-Autopsy

Sunquest

PowerPath

Sunquest

PowerPath

N/A Sunquest

CoPath

Sunquest

CoPath

Sunquest

CoPath

Blood Bank &

Transfusion Medicine-Donor Center-Blood Bank

Mediware

-Lifetrak

-HCLL

Sunquest 7.1 BTS Module

Sunquest

v6.3 for Cell Therapies only (limited use).

Use BB at BWH

Mediware

-Lifetrak

-HCLL



Tissue Typing G4 N/A N/A – send specimens to

BWH

mTilda N/A N/A

Reference Lab Mayo

LIS Systems Across PHS (2016)Lab Impacts: Adapting to Enterprise Information Systems

(Epic go live April 2016)

Homegrown system Enterprise system

Ordering favorites Not permitted Allowed

Order sets Reviewed by lab Uncommonly reviewed by lab

Collection process Simple (but manual) Complex (but electronic)

Menu size Limited (95%) Most tests available (99%)

Lab test search Provides decision support, CDS visible when searching

Search capabilities primitive, does not store search results or provide visible CDS when searching

Decision support availability Custom, fast, not requiring programming

Extensive possibilities but requires many levels of approvals, implementation complex

In lab processing Manual steps, slow Rapid, efficient

Moving from Reports to Real-time, Interactive Dashboards

• Allow access to variety of lab and support staff to track key metrics for EHR usage, efficiency, and quality and safety• Much of data is downloadable into Excel for process improvement projects• Limits number of custom reports needed

Lab Hospital Metric

Percentage of samples received in core and microbiology labs that were ordered in Epic

93.8%

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

% Epic All Patients

Week 4 details:Inpatient = 96.1%Emergency room = 97.5%Outpatient = 91.1%

9/5/2016

24

8 PM 10 PM

80/3March 23

April 26

Outpatient Specimen Processing Changes

April Post-Epic (last 2 weeks): Collect to Receive = 153 min

March Pre-Epic: Collect to Receive = 237 min

Basics first: Ensure your EHR/LIS queues are worked

• EMR and LIS interfaces have numerous error queues that need a technical and sometimes a clinical eye on a periodic basis

• Numerous reasons for results failing to post, posting in wrong location, or failing to bill properly

• Failure to ensure that these are being reviewed in a timely manner can impair clinical care and billing

Application ReportsApplication Reports

Epic System and Reporting Tools

Clarity ReportsClarity Reports

Chronicles

(PRD)

Clarity

Data mart Data mart

Reporting

ShadowETL Process

Data mart

Hyperspace

Reporting WorkbenchReporting Workbench

• Real time & small data set

• More flexibility in customization

• Actionable to patient record

• Refreshed every night

• Analytical trending data

• Highly customizable

• Quick, real time data

• Little flexibility in customization

• Actionable to access patient record

Daily report of all Epic orders including problem list, diagnosis, order source (order set or search)r

Daily monitoring needed to assess for utilization issues

Key Fields for an EMR CPOE Report

• Ordering provider and department• Origin of order (Order set, facility list, personal preference list, database search)• EHR test code• Order number sent to LIS (to tie order to result)• Order attributes (Future/Standing, STAT/routine, etc)• Responses to questions asked during order entry• Provider comments

A few examples how this report can be used…

9/5/2016

25

If it’s on the menu it will be ordered… RBC Folate

• With our Epic daily report

we were quickly able to localize the orders to a single enterprise anemia panel• We first swapped out RBC folate for serum folate and later removed the RBC folate from the menu

Limiting Options is Often the Best Approach: SPEP with Immunofixation

• Allowing clinicians to order SPEP + immunofixation led to large increase in lab workflow for a highly manual test• Removed test from menu and only allowed clinicians to order SPEP with reflex immunofixation• Used an automated tool to change > 100 personal preference lists to swap out inappropriate order for preferred order • We are back to our pre-Epic baseline now.

Tracking Miscellaneous Test Requests

• Prevent workarounds that are not easily trackable (e.g. paper requisitions)• Provides data for determining if additions to menu are needed

Regular Review of Misc Test Report for Trends and Outliers

Appropriate

WHOLE EXOME SEQUENCING

MATERNAL CELL CONTAMINATION FOR SNP MICROARRAY

CFTR SEQUENCING, REFLEX TO DEL/DUP IF NEGATIVE

ZIKA

Already built

MATERNI T21 PLUS

BORRELIA MIYAMOTOI AB

SERUM IRON, TIBC.

BUPRENORPHINE LEVEL

Don't build

BABESIA PCR

• 0.5% of orders (20/4,000 per day) are Misc Tests• Possible outcomes of review: Build new tests, educate providers, or keep as misc test

9/5/2016

26

Summary

• Laboratory patient safety and efficiency starts in the EHR

• Standardization of Lab Information systems is needed to lower complexity of upstream and downstream processes

– Ordering, interfaces with EHR, interfaces with reference labs, collection systems, test codes

• Detailed monitoring of lab results and orders provides needed information for process improvement and utilization control

Thanks!

flexner report (2010) data analytics in the clinical · data analytics in the clinical laboratory...

Documents