# Root Cause Analysis

Τεχνίτη Νοημοσύνη και Ρομποτική

7 Νοε 2013 (πριν από 4 χρόνια και 6 μήνες)

112 εμφανίσεις

Root Cause Analysis

Farrokh Alemi, Ph.D.

Jee Vang

Definitions

Root cause analysis is a process for
identifying the causes that underlie
variation in performance, including
the occurrence or possible
occurrence of a sentinel event.

Sentinel event is a major adverse
event that could have prevented (e.g.
wrong side surgery)

Conducting Root Cause Analysis

Before a sentinel event occurs, an
investigative team is organized.

When a sentinel event is reported, the people
closest to the incidence are asked to record
facts (not accusations) about the event.

The investigative team meets and
brainstorms:

potential causes for the incidence

key constraints that if they were in place would
have prevented the incidence.

Causes are organized into direct and root
causes.

A flow chart is organized showing the direct

Analysis validated by checking assumptions
and accuracy of predictions

Examples

Investigation of eye splash and
needle
-
stick incidents from an HIV
-
positive donor on an intensive care
unit
using root cause analysis

The Veterans Affairs

root cause
analysis system in action.

Root cause analysis
in perinatal care
.

Root
-
cause analysis of an
airway
filter occlusion.

Definitions Continued

Bayesian networks transfer probability
calculus into a Directed Acyclical Graph
and vice versa.

A Directed Acyclical Graph is directed
because each arc has a direction

The node at the end of the arrow is
understood as the cause of the node at

It is acyclic because there is no path
starting with any node and leading back to
itself.

Probabilities

Conditional independence implies a
specific root cause graph & vice
versa

Probability calculations are based on
assumptions of conditional
independence and vice versa

Conditional

Dependence

Root Cause

Graph

Probability

Calculus

Conditional Independence in
Serial Graph

Root

cause

Sentinel

event

Direct

cause

Conditional Independence in
Diverging Graph

Cause

Effect

Effect

Weight

gain

Diabetes

High

blood

pressure

Conditional Independence in
Complex Graphs

Any two nodes with a direct connection
are dependent

Any two nodes without a direct connection
are independent if and only if:

Either serial or diverging

Not converging

If condition is removed, the directed link
between root cause and sentinel event is lost

Assumptions of conditional independence
can be verified by asking the expert or
checking against objective data

Identify Conditional
Independencies in the Graph

Prediction from Root Causes

Use Bayes formula and Total Probability
formula:

Use software:

page

Double click to self extract to directory Netica

U
N
C
U
U
U
n
n
C
P
C
P
C
P
C
C
C
S
P
C
C
C
S
P
)
(
)...
(
)
(
)
,...,
,
|
(
)
,...,
,
|
(
2
1
2
1
2
1
Netica

Create a New Network

Click on this &

click into white

space

Click on

this, click

on start,

click on

end

Double click on a node

Enter description with no spaces

Double click on node

Select Table

Enter 100 times marginal probability,
click for the “Missing probabilities”
button for the system to calculate 1
minus marginal probability

Enter marginal probability of poor
training as 12 standing for 12%

Recalculates

Remaining

probabilities

Double click on the node

Select table

Enter 100 times probability of effect
given the cause

Enter data for each condition. When
conditions change, probabilities cannot
be calculated from previous data

Select the button for calculating
remaining probabilities

Entering Probability of Not
Following Markings Given Poor or
Good Training

Calculates

remaining

probabilities

Enter Conditional Probabilities for
All Combined Direct Causes

Conditions

Probability of
wrong side
surgery given
conditions

Patient
provided
wrong
information

Surgeon did
not follow
markings

Nurse marked
patient wrong

True

True

True

0.75

True

True

False

0.75

True

False

True

0.70

True

False

False

0.60

False

True

True

0.75

False

True

False

0.70

False

False

True

0.30

False

False

False

0.01

Compile the Graph

Making Predictions

Select a node

Select the condition that is true

Read off probability of other nodes

Predict sentinel event from combination
of root causes

Predict most likely cause from
observed sentinel event

Estimate prevalence of root causes
from observed direct causes

Predicting Sentinel Event With No
Information on Causes

Predicting Sentinel Event with
Three Observed Causes

Predicting Prevalence of Fatigued
Nurse if Patient is Marked Wrong

Selecting Most Likely Cause of
Sentinel Event

Discussion

Estimating the probabilities can verify
if assumptions are reasonable,
conclusions fit observed frequencies,
and help select most likely cause.

JCAHO reports some conditional
probabilities

Experts estimates are accurate if

brief training in conditional probabilities

Provided with available objective data

Allowed to discuss their different
estimates

Take Home Lesson

Question the obvious.