Towards Patient Safety in Closed-Loop Medical Device Systems Authors David Arney, Miroslav Pajic, Julian Goldman, Insup Lee, Rahul Mangharam, Oleg Sokolsky Presenter Velin Dimitrov
Nurses and other clinicians deal with a multitude of tasks They need to quickly assess dangerous situations and take preventive action Delays are life-threatening Adding autonomy to medical devices will relieve the burden on nurses Need
The clinician “closes the loop” Alarm sounds when sensor passes threshold Clinician arrives to assess the situation Clinician must first acquaint themselves Clinician takes action Typical Clinical Setting
What if devices themselves could sense common fault/dangerous conditions and respond more quickly than a clinician could? Human caregiver will always be in the loop! Proposal
ControllerPlantSensors Bridge to Control Systems
Formal, timed automata based model UPPAAL tool Exhaustively test system behaviors in failure mode Timing constraints from dynamic model Detailed, informal model Simulink/MATLAB Captures dynamics of human/device interaction Two Models
MD PnP and ICE Architecture Case Study
Patient Control Analgesia (PCA) pump Provide pain meds to patient Customized dosing Programmed limits on how many doses can be delivered Clinical Use Case
Mis-programming Overestimation of maximum dose Wrong concentration Accidental pressing of button These failures cannot be currently avoided Modes of Failure
System
Control Loop
Programmed by caregiver Normal rate of infusion Increased rate of bolus Bolus total duration Drug limit Built in sensors to catch device faults Network interface for status No pumps can currently accept control sigs PCA Pump
Measure SpO2 and HR Finger clip sensor Ratio of IR to red light Amplitude Pulse Oximeter
Pain, Pain-controlled, Overmedicated Critical Region Overdose, Respiratory distress SpO2 <70%, HR <11.5 bpm Drug level is a linear mapping to HR and SpO2 in this model Patient Model
Decides when to stop pump to keep patient out of critical region Clinical application script (CAS) Alarming Region Sp02 <90% or HR <57 bpm Notify caregiver – Alarming condition Supervisor Model
Communication Structure
Will the system function correctly? Finding faults and recovering gracefully Verification and Validation
Formal Model –Pulse Oximeter
Formal Model – PCA Pump
Formal Model - Supervisor
Formal Model - Patient
Formal Model - Network
Check that the pain eventually goes up in the model Check that the pump is stopped in the alarming condition Verifying Safety Properties
Used to determine the timing/rate parameters that make the system safe Models patient dynamics, network delays, pump delays Informal, Detail Simulink Models
System
Patient Model
PCA pump will always be stopped before we reach critical condition Safety Requirement
Variables
Finding t_crit
alpha is 0.001s^-1 Half life of drug is 11.5 minutes For H1 = 90% and H2 = 70% Tcrit = 26.8 minutes Comparing Time Delays to tcrit
Supervisor control algorithm and pump design must maintain open-loop stability Essentially adding capability to limit given dose per command from the supervisor – activation command Network Delay Tolerance
Disregard button pressed for tdel time units t_del must be less than t_safe for this to work t_del
t_safe t_safe must satisfy the following condition
dl_max = 100, Hdl_2 = 85.71, Hdl_1 = 28.57, dl_cur = 20 This corresponds to alarm/critial cond t_safe = 1723 sec t_safe
Supervisor dynamically sets max drug level Retrofittable Solution
Alaris 8210 SpO2 mdoule connects to Alaris 8000 pump controller Tightly integrated system from single vendor Need good model that captures whole process of drug delivery Pharmacokinetic models are not sufficient Related and Future Work