Gpt buchman

11

Better Health, Better Care, Lower Cost :How telehealth and real-time analytics can help critical care

achieve this triple aim.

Emory Critical Care Center

Tim Buchman

22 March 2014

Disclosures: None Relevant to the Presentation

22

Why monitor? :“Situation Awareness”

33

Mica Endsley’s Original ConceptionHuman Factors 37: 32-64 (1995)

44

SA involves more than “more data”

55

SA “on the road”

66

• Present all information in readily interpretable form, much as a GPS receiver takes data from satellites and creates situational awareness to provide a map back to health

Desiderata

77

Situation Awareness: Why does this “feel right”?

1. The perception of the data2. The comprehension of its meaning3. The projection of that understanding into the future in

order to anticipate what might happen

88

This is NOT SA…

99

Because excess, uncorrelated data constitute distractions…

1010

ICUs, Present Day

Loss of situational awareness is easy and common

1111

Staff cannot absorb more data. Really.

In today’s ICU, there is too much opportunity for error

Do distractions matter in critical care?An experimental study

Task: Alarm and vent checks Distraction:”I’m ready for handover!”

Miss rate, 25%

1313

The Four V’s of DataChallenge National Security Medicine Need

Volume

“We are... swimming in sensors... and drowning in data"

• Medical literature doubling every 19 years• Torrent of patient data

•Management of large data •Transform data to information

VelocityDecision timelines range from days to seconds

Decision timelines range from days to seconds

Rapid extraction and presentation

Variety

Range of data types: imagery, video, signals, seismic data, field reports, informants, news reports

Physiology, lab tests, physician notes, interventions, patient history

• Data association • Information representation • Data provenance

Veracity

Military operations, targeting, collateral damage, rules of engagement

Diagnosis & treatment of patients, life & death decisions, side effects, complications, malpractice concerns

High-confidence decisions: Costs of mistakes are high

1414

“In the moment”—what is the current physiologic status of my patient?

“Flowing data”—What is the trajectory of my patient?

Data (4Vs)-> Monitoring-> Situation Awareness

1515

Challenge Medicine Need

Patterned Biology, and especially pathobiology , is not random. The state space is “lumpy”. Treatments are aimed at lumps.

Not all patterns are evident to clinicians. Management of large data requires meaningful pattern detection.

Personalized There are three time scales that influence personalization: •Inherited aspects (“forever”); •chronic aspects (acquired, “allostasis”); •acute aspects (immediate threats, “homeostasis”)

Data often convolve all three time scales. Knowing the patient’s set-points and dynamics around the set points matters.

Predictive Prediction horizons related to the time scales, e.g.•Lifetime risk for cancer•Obesity risk related to environmental stress•Arrhythmia risk due to electrolyte disturbance

All three horizons require not only situation awareness but also a mechanism of alerting when the risks change. By extension, risk-management implies ongoing “what-if” scenarios.

The 3 P’s that Matter to Health Care

1977

(single dimension)

1977

(multidimension)

1818

Does this matter?

Yes, it does. An example…Duration of hypotension before initiation of effective antimicrobial therapy = critical determinant of survival, so knowing a single parameter contributing to the state affectsdecision-making

Kumar A, et al. Crit Care Med 2006;34:1589

State= “sepsis”

1986

1986 1969

time→

2

Simulation Evaluation of an Enhanced Bedside Monitor Display for Patients With Sepsis.Giuliano, Karen; RN, PhD; Johannessen, Ann; RN, MSN; Hernandez, Cheri; RN, MSN; EdD, CCRN

AACN Advanced Critical Care. 21(1):24-33, January/March 2010.DOI: 10.1097/NCI.0b013e3181bc8683

Simulation Evaluation of an Enhanced Bedside Monitor Display for Patients With Sepsis.Giuliano, Karen; RN, PhD; Johannessen, Ann; RN, MSN; Hernandez, Cheri; RN, MSN; EdD, CCRN

AACN Advanced Critical Care. 21(1):24-33, January/March 2010.DOI: 10.1097/NCI.0b013e3181bc8683

2525

Does this matter?

One of our ICUs, 3 years ago

2727

“Patterned, Personalized, Predictive”

2828

●Physiologic time series–Heart (EKG)

–Vasculature (Blood Pressure)

– Lungs (CO2)

–Brain (EEG)

–…

Detecting patterns at the bedside

Beat-to-beat heart rate

heartnt heart

nt

1

time, sec

ECG II,mV

Goldberger AL, et al. Proc. Nat. Acad. Sci. USA (2002) 99 Supp1:2466

Which is the healthy pattern?

Heart Failure



Heart Failure Heart Failure




Atrial Fibrillation




Normal Atrial Fibrillation



3434

●Nonstationarity

– Statistics change with time

●Nonlinearity

– Components interact in unexpected ways ( “cross-talk” )

●Multiscale Organization

– Fluctuations/structures typically have fractal organization

Patterns of health->Inferences about “not” health

Healthy Dynamics

3535

What “not health” looks like

Goldberger, Peng, Costa. Nature 1999; 399:461; Phys Rev Lett 2002; 89 : 068102

Healthy dynamics are poised between

too much order and total randomness

The breakdown “data patterns” are similar in various organ systems

3636

“Not health” : infection (sepsis)

• Two similarly septic patients

• First 24 hr of data shown

• During the second 24 hr, the patient on the right developed multiple organ failure and died on day 12.

Pontet J, et al, J. Critical Care (2003) 18:156

22% reduction in mortality!

If data-driven prediction was a drug in this setting, that 22% reduction in mortality would make it a BLOCKBUSTER

3838

Reengineering Critical Care

Patients and Conditions

Population Specification Populations

Care Path Development Fully Specified Care Processes and Protocols

Current CareWorkflow

ModificationDelegation,Algorithms

Situation Awareness,Response

Caregiver and Patient Activation

Low Efficiency and Reliability High

• Recognize physiologic decompensation as it occurs

• Classify decompensation by actionable mechanism

• Mitigate decompensation by reversal of cause and supportive treatment


• Harvest data in motion

• Real-time analytics

• Intuitive display

• Reliable interventions


Center for Critical Care

Data in Motion and Real-Time ICU Analytics

Testing Novel Analytics

Synchrography

π R

adia

ns

•Situation Awareness:Current State

Philips eICU

ECCC

Coarse data

Fine data

Quasi-real-time display and analysis of physiologic data: architecture that we are currently using

Numerics and Waveforms (240 Hz)

~ 10 sec latency


Architecture Example

Filter ECG data

RR Beat Detector

SampEn COSEn LDS

Database

BedMasterEx

Filter ICU Beds


ECG with beat detection

Analytics, etc.MIT-BIH: 12 beats q30 min for 24 hours

400 600 800 1000 1200

0

100

200

300

AF NSR CHF III and IV CHF I and II

mea

n of

the

stan

dard

dev

iatio

n

mean RR interval


Coefficient of sample entropy (COSEn)• An entropy metric

optimized to detect atrial fibrillation in very short records.

• It has ROC area 0.98 for detecting AF in 12-beat records.

0 20 40 60 80 100-4

-3

-2

-1

0

AF male AF female NSR male NSR female

CO

SE

n

Age (years)

Lake and Moorman, Am J Physiol, 2011Demazumder et al, Circulation 2013


Real-time COSEn/AF Example

5252

Making the tools work: the eICU platform

Better Health(outcomes that matter to patients and families)Better Care(High-reliability and evidence based)

Lower Costs (Optimal configuration of people and materials)

Right Care, Right Now, Every Time

Execution LayerStrategy

Workforce

Operations Plan

Ensuring that every test, drug, and

procedure add value to care

Event driven Intervention

1. Multiple event initiation triggers: such as requests from site (eLert button); admission/transfer event; detection of deterioration or collapse; advisory from another eICU staffer1. Consistent (normative) behaviors2. Verification that outcomes are achievedProcesses Matter

1. Bundles are “DO-LISTS”2. Standard list-driven responses to common care

challenges in critical care 3. Responses are also “DO-LISTS”4. eRN and eMD are PARTNERS in verifying

adherence to standard bundles: DO-LISTS completed

5. eRN and eMD are PARTNERS supporting standard responses to common situations. DO-LISTS completed

6. eICU collaborates with ICU staff to verify desired results are driven by standard bundles and interventions

7. Scheduled e-rounding for initiation and adherence to “bundles”

8. Two-person e Staff confirmation of DO-LISTS completion

9. Remote support by eICU for bundle/response order sets.

Value derives from what we do, making a difference

1. Debridement of drug lists2. Elimination of unnecessary standing orders3. Conversion to less expensive choice or route4. Avoidance of complications (drug interactions)

ECCC-eICU

Driver Diagram

Key Drivers Interventions

5454

●A lot of technology, rivers of data, lots of expense → opportunities to create and deliver value

●‘In the moment descriptions’ of ‘where the patient is’ would be very helpful (“situation awareness”)

●Predictive analytics to drive towards treatment goals would be very helpful

●Predictive analytics that fail (patients off the predicted trajectory) even more important

Takehomes

5555Questions?