BINF 7510 Clinical Decision making and Decision Analysis

1
BINF 7510Clinical Decision making and Decision
Analysis
Clinical Decision Support Systems Part II
2

• Hospital Based Decision Support
• These type of systems provide clinical care in
  hospital setting known as HIS( hospital
  information Systems). We will discuss a system
  located at LDS hospital at Salt Lake City,
  developed by the department of medical
  Informatics of the university of Utah.
• The system consists of an integrated database and
  it is a frame based medical decision support
  system.
• The system also includes support programs for
  hospital and department administrative functions.
  Relevant tools for maintaining and upgrading the
  information, knowledge and database are also
  included.
• The integrated clinical database contains a
  variety of patient data kept on-line during
  patients stay in the hospital.

3

• Database can be accessed by the healthcare
  professionals the terminals throughout the
  hospital.
• Terminals allow the entry of pertinent clinical
  data into the HELP ( Health Evaluation through
  Logical Processes) system by all personnel who
  are involved in patient health care.
• Typical information consists of ( partial list)
• Medication prescribed
• -Allergies -Physical examination
• -Blood gases -Admit/discharge info.
• -Electrocardiogram -X-ray findings
• -Demographic information -Dietary information
• -Cardiac data -Surgical procedures
• -Biopsy results -Procedures reports

4
-Hematology -Respiratory notes -Pulmonary
functions -Microbiological data -Nursing
data -Pathology department data Categories of
Decision Support Technologies 1. Processes which
respond to clinical data by issuing an alert. 2.
Programs that respond to recorded decision to
alter care by critiquing the decisions and
proposing alternate care - as appropriate. 3.
Applications that respond to a request by
decision maker by suggesting a set of diagnosis
of therapeutic maneuvers fitted to patients
needs. 4. Retrospective quality assurance
applications where clinical data are abstracted
from patients records and decisions about the
5
quality of care are made and fed back to care
providers. Alerting Systems These systems are
programs that function continuously, monitoring
selected clinical data as it is stored in
patients electronic records. The test programs
are designed to test specific types of standard
data against predefined criteria. If the data
exceed the predefined values, the system alerts
medical personnel by beeping or displays. Timing
and the character of the message vary with
altering goals. For example, one of the
sub-systems implemented on the HELP system
monitors common laboratory test results. After
detecting any abnormal conditions, life
threatening conditions or deviations of data from
the set criteria,the systems issues an alert
signal on the terminal and to the care givers.
6
The HELP system captures results from clinical
laboratory tests through an interface to a
dedicated laboratory information system ( LIS.
Results are collected by and returned to the HELP
system for storage in the clinical record as soon
as they are collected and validated in the
LIS. These results are reviewed by personnel
engaged in the in-patient care. For example, the
physician may visit only at certain times. Under
such circumstances, the abnormalities in the
laboratory test results may not receive timely
attention.. HEP system provides an alert, if
conditions are abnormal and warrants for an
immediate attention. Two approaches are used for
delivering the alerts 1. A fleshing yellow
light was activated by the system. It continued
to flesh until the alert is reviewed and
acknowledged on the computer terminal.
7
2. Whenever any care giver ( physician) entered
the program to review patients laboratory
results, any unacknowledged alerts were
immediately displayed along with the relevant
data that had triggered the alert. Results of
this type of alert were tested by the system in
three ways. These are First, the appropriateness
of the treatment was evaluated. The alerting
system would suggest appropriate therapy for
condition showing abnormalities ( say involving
Na, K and glucose levels). Second, the time
spent in the life threatening condition with and
without the alert. Finally, the hospital length
of stay was examined. A significant improvement
in this parameter was noted for the patient with
above abnormalities.
8
This type of decision support intervention is
becoming increasing common as hospital
information systems evolve. In the inpatient
environment, where the severity of illness is
steadily increasing, there is a strong potential
for better alert systems. For the HELP system,
monitoring Na, K or glucose levels, the
following alerting logic was used Hyponatramia
(NAL) Na lt 120 m Eq/l Falling sodium(NAF) Na
has fallen 15 m Eq/l in 24 hours and Na lt
130 m Eq/l Hypernatramia (NAH) Na gt 155 m
Eq/l Hypokalemia(KL) K lt 2.7 m Eq/l Falling
Pot. K fallen 1 m Eq/l in 24 hours and
K lt 3.2 m Eq/l
9
Hypokalemia K gt 6.0 m Eq/l Hypoglycemia (
GL) gl lt 45 mg Hyperglycemia ( GH) glgt 500
mg CRITIQUING SYSTEMS Critiquing process in the
HELP system begin functioning when when an order
for a medical intervention is entered into the
system. Such systems typically respond by
evaluating an order and either pointing out
disparities between the order and the internal
definition of proper care or by proposing an
alternate therapeutic approach. Example ( order
for a blood product)- blood product must be
ordered and administered with care - Diseases
that can be transmitted during transfusion could
be deadly.
10
- Short life span of blood products make this a
scarce resource to be used sparingly. - There are
certain criteria/guidelines to be satisfied and
system carefully monitor those. Computer program
at the LDS hospital was specifically designed to
manage ordering of blood transfusions and to
assist in ensuring the compliance. It was
essential that all blood products are to be
entered through computer system. Embedded into
the program is a critiquing tool designed to
ascertain the reason for every transfusion and
compare the reasons to specific
guidelines/criteria. For example when an order is
made for packed red-blood-cells, the following
logic is applied
11
Hemoglobin lt12g/dl or hemotocrit lt 35 if age
?35 years Hemoglobin lt10g/dl or hemotocrit lt 30
if age lt 35 years Oxygen sat (SaO2) lt
95 Active bleeding Blood loss gt 500
ml Systolic blood pressure lt100mm Hg or
Heart-rate gt 100 bpm Adult respiratory distress
syndrome ( ARDS) After going through this logic,
the following procedure is followed (i) The
physician is shown the blood products ordered in
the last 24 hours for the patient in
question. (ii) Appropriate laboratory data are
displayed.
12

• (iii) Then, user chooses chooses specific blood
  products along with the number of units and
  priority.
• (iv) User is subsequently asked to document the
  reasons for the order.
• (v) A list of reasons, specific to blood product
  chosen, is displayed and the user chooses
  appropriate rationale for the intervention.
• (vi) Computer applies the stored criteria and
  determines whether the order meets the hospitals
  guidelines.
• If the guidelines are met, the order is logged
  and the blood bank and nursing divisions are
  informed electronically and via computer printout
  also.
• If the set criteria is not met, the user is
  presented with a message stating the applicable
  criteria and relevant data. The physician may
  place the order or cancel. However, a further
  justification is required to override the system.

13

• One way of measuring the effectiveness of the
  critiquing system is to examine the frequency
  with which the process of ordering blood products
  is terminated as a result of this expert HELP
  system.
• During the six-months test period , the ordering
  program was entered and exited without a order
  for 677 times. This reflected a total of 12.9 of
  the total blood product uses.
• At least half of these cases did not order (
  terminated) blood products based on systems
  recommendations.
• The system relies heavily on the integrated
  database in HELP system. It accesses data from
• (i) admitting department (ii) clinical
  laboratory
• (iii) surgical scheduling (iv) blood bank and,
• (v) orders entered by nurses and physicians.

14
The program also responds to interventions chosen
by physicians by analyzing orders and if
appropriate, suggesting reasons to alter the
therapeutic plan. Critique can provide a series
of informational responses designed to assure
that the user is fully aware of both the status
of the patient and also the accepted guidelines
governing the usage. Physicians often appreciate
the ability of the automated ordering system to
give feedback on the proper dosing and care
protocols. Suggestion System The third category
of computer applications designed to support
medical decision-making is potentially the most
interactive procedure. The group of processes are
designed to react to requests for assistance.
These processes respond by making concrete
suggestions concerning which actions should be
taken next.
15
Clinicians would typically call-up a computer
screen, enter the requested query or data and
wait for the suggestion from these systems before
instituting a therapy. Unlike critiquing system,
the physician need not commit to an order before
program applies its stored medical logic.
Instead, the program conducts an interactive
session. It ask many questions during the
process, then reviews the relevant data, and
formulate a suggestion for an intervention based
on medical knowledge stored in its knowledge
base. One of the functions - Ventilator Therapy
System - has been used very actively since 1987.
The details are as follows LDS hospital sees a
large number of patients with respiratory
failures. One of the more difficult problem is
that of adult respiratory distress syndrome -
ARDS.
16

• This disease can complicate a number of other
  conditions including trauma, infectious diseases,
  and shock.
• The usual therapy include respiratory support
  while underlying pulmonary injury heals.
• The mortality rate in this case is close to 50.
• For ARDS patients severe hypoxemia - this
  rate climbs to 90.
• A set of protocols were developed for using CO2
  removal . Since protocols were complex, computer
  based system was developed.machine.
• ECCO2R ( extra-corporal CO2 removal) system was
  tested on 40 critical patients. 21 with ECCO2R
  and 19 with conventional therapy. There were 7
  survivors of ECCO2R group and 8 in conventional
  method.

17
Although, there was not a significant
achievement, however, everyone accepted that the
quality and uniformity of the care provided by
computer protocols has resulted in a significant
improvement in patient outcome. As a result, the
computerized monitoring system continued. After
some time, the results were published and are as
follows
18
DIAGNOSTIC DECISION SUPPORT WITH HELP SYSTEM One
of the greatest challenge for computerized
medical decision making systems is to participate
productively in the diagnostic process. Clinical
diagnostic decision support systems( CDDSS)
differ from decision support systems described so
far. - Decision support systems can draw
attention and can synthesize therapeutic
suggestions based on their knowledge base. - On
the other hand, the diagnostic process is a
preliminary step in making therapeutic
intervention. Diagnostic systems may require a
system with different goals, interfaces, and
decision algorithms than the applications
previously described. A number of applications
residing in the HELP system can, through the use
of various diagnostic strategies, affect patient
care. Some of these applications are 1.
Evaluate patient data to detect adverse adverse
drug events.
19
2. Tools that recognize nosocomial infections. 3.
Computerized assistant that informs and advises
physicians as they undertake a complex task of
determining how to treat a patient with a
possible infection. 1. Adverse Drug Events(
ADEs) These are defined as response to drug
which is noxious, unintended and which occurs at
doses normally used in man for prophylaxis,
diagnosis, or therapy of disease. ADEs can range
in severity from drowsiness or nausea to
anaphylaxis and death. In USA, the drug related
morbidity and mortality cost over 136 billion
per year. ADE subsystem continuously monitors
patient for ADE by inspecting patient data
entered at the bedside for any signs of rash,
respiratory rate, heart rate, hearing, mental
status, seizure, anaphylaxis, diarrhea, and fever.
20

• In addition, data from clinical laboratory,
  pharmacy, and medications, charting applications
  are analyzed to determine any symptoms of ADEs.
• The system evaluates all of the patients in the
  hospital and generates a daily computer reports
  indicating which patients are possible ADE
  victims.
• A clinical pharmacist follows-up on these
  patients and completes the evaluation using a
  verification program.
• This program provides a consistent method for
  completing the diagnostic process by giving a
  possible score on the scale of 1 - 10.
• The physicians caring for each patient are
  notified of the confirmed ADEs.
• For May 1988 to May 1989, a total of 401 cases of
  adverse reaction cases were detected

21

• Additional effect of this program appears to be a
  reduction in the number of serious ADEs.
• The system used early detection technique to
  notify the physicians of possible cases of ADEs.
  The number of severe cases reduced from 41 in
  1991 to 12-15 in subsequent years.
• Investigators also noted that the length of
  hospital stay of patients with ADE was increased
  by 1.91 days and cost by 2,262. Computer
  system could reduce this stay and saved cost as
  well. The quality of the care also improved
  significantly.
• These tools leverage the fact that the majority
  of the data necessary for their function was
  available in HELPs integrated database.
• They have illustrated the potential for
  computerized diagnostic applications to impact
  patient care by providing good clinical attention
  and saving life of many patients.

22
The system used for this application is basically
ruled based. Nosocomial Infections This
sub-system is designed to recognize nosocomial
or hospital acquired infections. The process of
detecting nosocomial infections serve a
recognized purpose. Control measure based on this
information are believed to be important in
interrupting the spread of hospital acquired
infections. Evidence suggests that intensive
surveillance programs may be linked to reduced
rate of infection. However this process can be
expansive. Traditional techniques require
infection control personnel to screen manually
all appropriate patients on a routine basis. The
computerized surveillance system used in LDS
Hospital relies on data from variety of sources
to diagnose nosocomial infections.
23
Information from microbiology laboratory, nurse
charting, chemistry lab, admitting office,
surgery, pharmacy,radiology and respiratory
therapy are used. Once each day a report is
produced detailing computers suggestions. This
report can be used to follow-up actions for
patients for whom there is evidence of
nosocomial infection. in a case study 217
patients were identified out of 4,679 patients
over 2 months study ( 182 cases were identified
by computer and overlapping 145 by traditional
means). Out of this 155 were confirmed to have
nosocomial infections. Computers sensitivity
was 90 with a false positive rate of 23, while
the infection control physician demonstrated 76
sensitivity and a false positive rate of 19.
When number of hours required to each approach
were estimated, computer based approach was mare
than twice as efficient as the
24
Manual approach. The nosocomial infection tool,
like ADE recognition system, uses boolean logic
in a relatively simple diagnostic process. As an
enhanced tool, one of the functions of this
sub-system was to predict which patients were
likely to acquire the infection. the tool is
based on some decision algorithms. Data from
patients with infections acquired in hospital
were combined with data from a control set of
patients, and a group of statistical programs
were used to identify risk factors. Logistic
regression using these risk factors was used in
the development of tools that could estimate the
risk of hospital acquired infection for
inpatients. The resulting system is capable of
predicting 63 of the population who are
ultimately affected.
25

• A computer based system has also been used to
  prescribed antibiotics. During a 7-year study, a
  fraction of patients who received antibiotics
  increased each year.
• However, the total cost of antibiotic decreased
  from 15 to 13 of the total drug
  expenditures.(fewer doses were required)
• The system also monitors the subset of surgical
  procedures for which prphylactic antibiotics are
  recommended ( ie, like hip replacement).
• For these procedures, antibiotics are often
  missed or given at a wrong time. Sometimes, once
  started, the antibiotics are not discontinued at
  the recommended time. In the absence of
  infection, a small number of doses are required.
• Based on computer,s reminder, the number of
  patients who were given this antibiotic increased
  from 40 ( of those who needed it )

26
to over 99. Average number of doses decreased
from 19 to 5.3 by the end of the 7-year study
period. This suggests that the computer based
systems can improve antibiotic use and decrease
chances of infection. Antibiotic
Assistant Infectious diseases department at LDS
has also developed a tool to help clinicians make
an informed decision concerning administration of
antibiotics. The antibiotic assistant provides
three basic services 1. It assembles relevant
data for physicians, so that they can determine
if a patient is infected and what intervention is
appropriate - information, such as renal
function, allergies, and temperature patterns are
presented.
27

• 2. System suggests a course of therapy for the
  patient.
• 3. The program allows a clinician to review
  hospital experience for the past 6 months and 5
  years (relevant to infections/ interventions).
• There is also a provision of system explanation,
  if requested.
• The data and effectiveness of antibiotics are
  analyzed on a monthly basis. The goal of the
  analysis to define the probability of each
  possible pathogen as a causative agent for a
  certain class of patients.
• Such information is regularly updated into the
  system and all the decision given by the system
  are based on decision rules. System is self
  learning type.
• The enhanced knowledge base is used by the
  antibiotic assistant program to make presumptive
  diagnosis of infectious oranisms

28
and to suggest treatments appropriate to these
organisms. The system always remains updated
through monthly updates in the knowledge base
through the learning mechanism. Computer based
help system also assists in Data collection -
basic questions are asked from the patients in
the beginning. Subsequently, based on the
response - more specific questions are asked and
the response is recorded in the database. Patient
data are compared with the diagnostic
frames. Additional questions could be asked by
the system, if necessary Assessing the quality of
X-ray reports - standards are created by getting
on X-ray assessed by 5 experts and recorded in
the database. Based on the knowledge, system can
examine the reports submitted to it and makes
comments and suggestions. System analyses
decisions periodically and the database is
updated.