Getting the Most from Clinical Data through Physiological Modelling

1
Getting the Most from Clinical Data through
Physiological Modelling Medical Decision
Support

• Bram Smith
• Stephen Rees, Toke Christensen,
• Dan Karbing, Steen Andreassen
• Center for Model-based Medical Decision Support,
  Aalborg University, Denmark

2
Introduction

• EXISTING TECHNOLOGY
• Clinical databases allowing easy, automated
  storage and retrieval of patient data.
• Medical equipment allowing data collection on a
  PC.
• Physiological models and decision support
  systems.
• BUT
• Doctors are still faced with interpreting large
  amounts of data to diagnose patients.
• PROPOSED SOLUTION
• An architecture that combines existing database
  technology with physiological models and decision
  support algorithms to assist clinicians in
  diagnosing and treating patients.

3
Systems architecture

• Clients on the architecture can be divided into 3
  types
• Inputs User inputs, or data taken automatically
  from medical equipment.
• Interpretation Uses equations and physiological
  models to expand knowledge about the patient.
• Decision support Uses decision support
  algorithms to assisting in choosing suitable
  treatment strategies.

Inputs Ventilator (Paw, Flow,) Gas analysis (O2,
CO2) Clinical monitor (ECG, HR,)
Interpretation Metabolic (VO2, VCO2,) Lung
(Shunt, V/Q,) Blood (Base excess, DPG,)
Decision support Monitoring Ventilator
control Glucose regulation
Database
Architecture is compartmentalised to allow
independent development of each client.
4
Input clients

• Many monitors allow data logging on a computer
  for automated data collection.
• The user interface also allows clinicians to add
  data that can not be logged automatically.

CO, MAP,
ECG, SaO2,
CO2, Vt,...
Vt, Paw,
Values, Events
Database
ALL data input is written to the database.
5
Interpretation clients

• Physiological models and more basic calculations
  are carried out on data in the database to
  determine more abstract measurements or patient
  condition and extend the knowledge of the
  patient.
• Some clients can be automatic, carrying out
  calculations when ever new data is available,
  while more complex clients may require user
  control.

Automatic
Requires user control
Body surface area
Cardiac Index
Oxygen Consumption
ALPE
Weight Height
BSA CO
FetO2, Vt,
VO2, CI,
Shunt, DPO2
BSA
CI
VO2
Database
6
Decision support clients

• The extended data set can be sorted and displayed
  in a way that assists clinicians in diagnosis and
  treatment selection. For example
• Analysing how a particular measurement has
  changed with time.
• Displaying only data that is relevant for the
  patients disorder.
• Methods of assisting in optimising treatment
  selection.

• Optimal
• PEEP,
• FiO2,
• Vt,

• History of
• Shunt,
• Deadspace,
• Insulin,

• Heart failure,
• ARDS,
• COPD,

Optimal insulin infusion
INVENT
Relevant information only
Hyperglycaemia
Plot history
PEEP, Vt,
Shunt, DPO2
Database
7
Implementation
CO, MAP, ITBV,
ECG, SaO2,
CO2, Vt, Paw,
O2, CO2
O2, CO2
CO2, Vt, Paw,
Decision Support e.g. ALPE
8
Conclusions

• A generic architecture has been implemented for
  development of new calculation methods,
  physiological models and decision support
  systems.
• The compartmental design means that clients can
  be developed and function independently, yet
  interact if possible to improve functionality
  (eg, cardiopulmonary interaction, VO2).
• This architecture presents a method for moving
  information systems from audit to clinical
  support tools, through
• Calculation of abstract representations of
  patient condition (e.g. cardiac index, shunt).
• Assistance in interpreting patient information
  and choosing optimal treatment strategies (e.g.
  optimising ventilator settings).