Critical Care Bioinformatics at Columbia University Medical Center J. Michael Schmidt, PhD Neurological Institute of New York Columbia University College.

Critical Care Bioinformatics atColumbia University Medical Center

J. Michael Schmidt, PhDNeurological Institute of New York

Columbia University College of Physicians and Surgeons

Disclosures

• CMA Microdialysis: Speaker’s Bureau

Clinical Needs for ICU Informatics System• Understand patient physiology• Evaluate effectiveness of treatments• Potential for automatic processing of data for

diagnosis and prognosis• Support clinical decision making

Also noteworthy…• Digital Note• Quality Assurance• Clinical Research support

Meeting the Need

3 problems + 1

•Data Collection•Data Interrogation •Data Analysis•Best use of data for clinical decision making

Columbia Neuro-ICU System

The

GOOD

• 18 bed ICU• Useful for threshold

based alarms and nursing workflow.

• Data only saved for 72 hours then gone.

• Not all device data can go into system

• Difficult to visualize data

• Saves parameter data every 5 seconds SQL database

• Saves visible waveform data

• Requires ~ 300 GB / year data storage

• Data from many bedside devices do not interface with the bedside monitor

• If they do interface, often ones want greater precision or all available parameters

Data Visualization

• Visualize data to track effects of interventions

• Can visualize physiologic relationships

• Multi-parameter event detection

• Collect parameter and waveform data

• Collecting data from devices that do not go into monitor

• Able to visualize data over extended periods of time

• Can elucidate relationships among different parameters

• Data stored permanently

The

BAD

ICP Osmolal-Serium 3% Saline Mannitol 20%

1/9/2007 1/10/2007 1/11/200700:00 00:00

0

50

100

mm

Hg

260

280

300

320

340

0

200

400

75 ml/hr

0

1000

2000

ICP Osmolal-Serium 3% Saline Mannitol 20%

1/9/2007 1/10/2007 1/11/200700:00 00:00

0

50

100

mm

Hg

260

280

300

320

340

0

200

400

75 ml/hr

0

1000

2000

• Labs available via copy and paste• Infusion pump data must be

manually entered• Can’t get it from EMR• Can’t get it from the pump

• Evaluate effectiveness of interventions

– Did the intervention have the desired effect?

– How quickly?– Does repeated administrations

work equally effectively?

• Tools to evaluate waveforms are lacking in our system.

• Phase analysis between MAP and ICP to determine autoregulation status.

• Heart rate variability analysis can be used to understand dysautonomia.

The

UGLY

Network Performance• Equinox boxes communicate with the

server using UDP• Must be in contact with server at

least once every 60 seconds else goes offline

• Must be rebooted to reestablish the connection

• Some will not go back

Network ReliabilityICU Pilot computers can’t always connect to

servers. • Server and computer operate on different

networks. Connection can fail (can’t ping server).

• Network Drive mapping fails– Map network drive– Setup ICU Pilot to use mapped network drive

for use as the database• Start SQL client, which fails to connect

sometimes, or gets corrupted and needs reinstall

Standard Informatics infrastructure for hospital level system

Maintenance: Server failure lead to corrupted database• Backup was not properly done• Used recovery tool to get data – currently running

scripts to recover the data will take 2 years to complete (need to write special program to make that faster). Who is going to do that?

Standard Informatics infrastructure for hospital level system

Server switch to Virtual server plus problems (OLTP versus OLAP)• I/O disk write time delays as databases got large

– Need separate system for data processing• Didn’t work the same.

– SQL didn’t load before bedmaster causing problems. • Processor didn’t necessarily always have enough resources to meet

demand

Summary of Columbia experience

1. Store data from the patient monitor forever in open SQL format

2. Can get data from devices that do not plug into the patient monitor or do so inadequately

3. Able to visualize relationships among parameters

1. Infusion pump, intervention, and lab data not in system

2. Tools not readily available to quantify the effectiveness of treatments

3. No tools to perform waveform analysis

1. Informatics infrastructure not robust.

2. Lead to:• Connectivity

failures• Backup systems

failed• Virtual server not

up to task.

?

• Other data systems not connected– Have to copy and paste in lab values– Infusion pump data must be manually entered

• Can’t get it from EMR• Can’t get it from the pump

• No good tools to evaluate effectiveness of interventions– Did the intervention have the desired effect?– How quickly?– Does repeated administrations work equally effectively?

• No tools for waveform analysis