Transcript Decision Support Systems

CLINICAL DECISION
SUPPORT
Dr. Saeed Shiry
& Clinical Decision SUPPORT
Road Map
Amirkabir University of Technology
Computer Engineering & Information Technology Department
A Case Study:
CLINICAL DECISION SUPPORT at LDS HOSPITAL


Medical decision-making requires the
clinician to apply accumulated knowledge to
a specific amount of patient information to
produce a result that may be a diagnosis,
prognosis, course of therapy, or the selection
of further tests.
Too often, the decisions are based on limited
knowledge, the information is incomplete or
imperfect, and the decisions must be made
during a limited period of time.
Clinical decision support


clinical decision support, has been defined as ‘‘any
computer program designed to help health
professionals make clinical decisions, deal with
medical data about patients or with the knowledge
of medicine necessary to interpret such data’’.
These DS tools can be classified into:
1.
2.
3.

tools for information management,
tools for focusing attention, and
tools for patient-specific consultation.
computer applications should identify and reduce
the rate of errors, inappropriate or inefficient
actions, and adverse events.
Architecture and key features of
HELP System
Electronic health record (EHR)





Most patient information is stored in the EHR.
 Some comes from applications that are part of the System, and
 Some comes from other applications
A number of medical devices also are interfaced to the System,
and patient vital signs, medication pump, and ventilator
information is stored automatically in the HER.
The clinical information is stored in a common database. The
transcribed dictations from x-rays, history and physical exams,
and other reports, and admission diagnoses are stored as freetext, whereas most of the EHR data are stored in a coded format.
Coded data are needed for the decision support process.
Each coded element in the database and the free-text data are
stored with an event time.
knowledge base



A second key feature of the System is a knowledge base that
contains thousands of medical logic modules (MLMs).
The MLMs contain medical logic that has been developed by
medical experts from different knowledge domains.
 Some contain simple rules to identify patients with elevated
potassium levels based on laboratory results;
 Others may contain complex logic and require patient information
from a number of data sources in the EHR.
Each of the MLMs contains two main parts.
 The first part identifies which data elements in the EHR are
needed for the logic.
 The second part contains the computer logic used to analyze the
data elements.
Time and Data Driver




The third key feature of the System is the ability to data and time drive
the knowledge base.
All data that are stored in the EHR pass though the data-driver, and
each data element is screened.
The first task the data-driver does is to make a copy of the data string
being stored into the EHR and to send that copy to the decision support
engine.
The decision support engine parses the data string into separate data
elements.



Each data element is then compared with the code tables to see if there are
any MLMs that should be activated based on that data element.
When the decision support engine runs the MLM, it retrieves additional
needed information from the EHR as specified in the MLM. If the logic
in the MLM generates an alert, another record is built and stored in the
alert file. This enables the system to continuously monitor patients.
The time-driver is simply a program on the system that checks a table
each minute to see if there are MLMs or other applications that should
be run.
Main functional processes of the
data-driver
Data Base






Another key feature of the System is that the data in the EHR are
never deleted.
As patients are discharged from the hospital and all billing is
completed, the patient record is moved to an archival EHR.
All clinical patient data from the HELP System since 1983 are
stored in the current or archival EHRs.
The archival storage of the EHR provides data that are often
analyzed and used to develop the medical logic contained in the
MLMs and has been essential for numerous retrospective
research studies.
Pharmacy System





The pharmacy application accesses patient data from the EHR to generate
alerts of potential adverse drug events; drug–drug, drug–allergy, drug–
laboratory, drug–disease, drug–dose, drug–diet, and drug–interval.
The application also generates prescription labels and patient drug profiles that
are used for unit dose dispensing.
The alerts are displayed to the pharmacists as they enter the hand-written
physician orders into the application. The pharmacists then inform the
physicians or nursing staff of the potential problems.
An evaluation of the pharmacy application showed that 5 percent of patients and
0.8 percent of drug orders generated alerts, and that physicians changed patient
therapy for 77 percent of the alerts.
The problem with this approach is that patients may receive the drugs before the
pharmacists enter the orders into the pharmacy application.

An approach to this problem is that physicians should not handwrite their orders but
enter them directly using provider order entry (POE) applications. That way the
physicians would immediately get the alerts and change the order before the drug is
administered.
Blood Gas Reports





The computerization of laboratory instruments could provide more
medical information and in less time.
However, this led tomedical staff being presented with large amounts of
patient data. Often this increase in information, although making the
medical decisions more accurate, required the medical staff to take
more time to gather and assimilate the pertinent information.
This situation resulted in computers being used to provide or assist in
the interpretation of laboratory test results.
This resulted in the reporting of blood gas results on the System to
automatically include the interpretations without any direct physician
interaction .
The accuracy of the computer interpretations was compared with the
interpretations of four pulmonary and three nephrology experts:


The results ranked the computer interpretations second among the experts.
A physician survey found that 80 percent of the blood gas interpretations
were helpful and 28 percent changed patient care.
Emergency Department Infection
Report








Thousands of patients visit emergency departments every day and based on their specific
clinical manifestations have specimens collected and sent for microbiology examination.
Often the laboratory tests are ordered only for precautionary purposes and the patients are
sent home. Every emergency department can relate stories about patients who were sent
home and subsequently had laboratory test results that contained important information
that was overlooked or not followed up.
The emergency department infection report is a simple printout that contains all the
microbiology and other infection-related test results for all the emergency department
patients during the past 10 days.
During each shift, a member of the emergency department staff is assigned to run the
program and examine the report for any new infection information.
When important information is found, the patients are contacted and given specific
instructions based on the test results.
Use of this information management tool resulted in contacting an average of two patients
each day and informing them that they need an antibiotic or that their previously prescribed
antibiotic needs to be changed.
This application demonstrates that the value of decision support applications is not
determined by the sophistication or complexity of the program(s) or database.
Over the past 20 years, the emergency department at LDS Hospital has changed
thousands of therapeutic decisions based on the information contained in the emergency
department infection reports.
Nurse Bedside Charting





Without the entry of medical data, tools for information
management and decision support could not function.
A major question is:

how, where, and when the information should be entered.
In some situations, direct data access is provided by interfaces to
medical devices, laboratory instruments, or other computer
systems.
The data provided from interfaces to other computer systems
often requires initial data entry by medical staff.
An important message of this chapter is that accurate and timely
computer decision support is dependent on the data available to
the decision logic in the knowledge base, hence the importance
of the information contained in the EHR.
Nurse Bedside Charting




An important source of medical information is that which is
acquired and documented by the nursing staff.
An electronic nurse-charting program was developed which
contained some decision logic that would alert the nurse when
patient information that was entered was out of range or
inappropriate.
nurse documentation done at the nurse station was generally
found to take place at the end of the shift, probably less
accurately, and thus was believed to have a reduced value for
decision support.
Study shows using the nurse-charting program at the bedside is
more accurate and useful.
Infectious Disease Monitor

This application identifies patients who have conditions that infection
control practitioners and infectious disease physicians want to be aware
of:







1) patients with hospital acquired infections,
2) patients with reportable diseases,
3) patients with antibiotic-resistant pathogens, and
4) patients with infections in sterile body sites.
The code for a microbiology test directs the decision support engine to
load MLMs that contain logic to identify pathogens based on the
specimen and/or body site.
Some MLMs contain logic to determine which infections need to be
reported to state and federal health departments whereas others
contain criteria for the identification of hospital-acquired or nosocomial
infections.
Thus, as patient information from microbiology culture results,
urinalyses, and chest X-rays are stored in the EHR, the data-driver
provides 24 hour and hospital-wide surveillance.
using the time-driver to notify
possible hospital-acquired infections
Prediction of hospital Infection




They wondered if they could use statistical methods to identify
patients at high risk of developing an infection in the hospital
before the infection onset.
A study database was created with 3,151 patients with hospitalacquired infections and 3,152 control patients. Stepwise logistic
regression was used to develop a predictive model for high-risk
patients based on 10 of 18 putative risk factors tested.
A computer program was activated each day to use an equation
based on the model to monitor all hospitalized patients and
create a computer printout of the high-risk patients.
During the first six months 78 percent of hospitalized infections
occurred in high-risk patients and 63 percent were predicted
before the documented onset of the infection.
Data Mining:
Adverse Drug Event Monitor



They began to question if a drug that cost a few
dollars less per day to use but caused a number of
ADEs was really less expensive than another drug
that caused only a few ADEs?
No one at the Hospital had any idea what the actual
ADE rate was, nor which drugs caused the ADEs.
MLMs were developed that monitored:




1) laboratory test results that could be indicative of a
possible ADE,
2) elevated serum drug levels,
3) the ordering of drugs that are commonly used to treat
ADEs, and
4) physiologic data that could signal possible ADEs
A New Version of the system


The new Web-based and fully integrated system has
allowed a pediatric intensivist to use the Internet and
access a child’s laboratory, medication, and ECG
information at a hospital 305 miles away and make a
life-saving diagnosis and therapeutic change.
The new anti-infective management program is
more accurate, with access to microbiology, chest xray and other patient information obtained at one
IHC facility before the patient is transferred to
another IHC hospital.
messages provided from the
medical decision support











The timing of data entry is critical. Patient information needs to be entered into the EHR as
soon as possible
Successful decision support applications are developed by a team consisting of clinical
domain experts providing the why and what needs to be done and the medical
informaticists providing the how.
Decision support should be integrated with the daily work processes of the medical staff
and occur at the appropriate point of patient care. Patient alerts should be sent directly to
the most appropriate people as soon as possible.
Decision support applications need to be tested for safety before they are made available
for general use. One bad experience can create barriers or restrictions for any future
applications.
Often large patient care improvement projects need to be broken down into smaller more
manageable processes.
The medical logic and rules need to be evidence based and match local processes of
patient care.
The logic and rules need to be periodically reviewed and updated as patient care and
technology change.
The applications must be easy to use and training should not be so difficult that patient
safety could be compromised.
Evaluation of medical decision support applications is often the hardest part.
The applications need to be cost effective and reasonable to implement and maintain in
order to gain administration support as well as clinical support.
Physician support of order entry is easier to get if all orders, laboratory, medication,
radiology, and so on, can be made at the same time using the same application.
Homework
One of the Following Papers from:


1.
2.
Clinical Decision Support Systems Theory and
Practice, Second Edition
Data Mining and Clinical Decision Support
Systems
Design and Implementation Issues