Transcript Document

Pediatric Knowledgebase (PKB) -- A Hospital-based, Decision Support System to Guide Pharmacotherapy
Jeffrey S. Barrett, Peter C. Adamson, Mahesh Narayan, Athena Zuppa Jeffrey M. Skolnik, John T. Mondick, Robin Norris, Kelly Wade,
Bhuvana Jayaraman, Dimple Patel, Erin Cummings, Kalpana Vijayakumar, Sundararajan Vijayakumar
Laboratory for Applied PK/PD, Division of Clinical Pharmacology, The Children’s Hospital of Philadelphia, PA
BACKGROUND
Pharmacotherapy is generally concerned with the safe and effective
management of drug administration. It implies an understanding of
drug pharmacokinetics (PK) and pharmacodynamics (PD) so that
individual dosing guidance, when necessary, can be provided to
optimize patient response within their individual therapeutic window.
Pediatric pharmacotherapy can be challenging due to developmental
changes that may alter drug kinetics, pathophysiologic differences that
may alter pharmacodynamics, disease etiologies that may be different
from adults, and other factors that may result in great variation in
safety and efficacy outcomes. The situation becomes more convoluted
when one considers critically ill children and the paucity of wellcontrolled pediatric clinical trials. This situation, despite the efforts of
the Food and Drug Administration and the US Congress, is not likely
to improve substantially due to the economic reality of the pediatric
market. Our long-term research goal is to facilitate the safe and
effective administration of drugs used in the treatment of children. Our
objective is to create a patient-based informatics system which
contains the relevant data concerning dosing of specific agents to
various pediatric subpopulations. What will distinguish our system
from existing commercial offerings is that we will employ analysis
and visualization tools, in some cases sophisticated models to forecast
patient outcomes to dosing scenarios. Our central hypothesis is that
dosing guidance can be improved when the caregiver responsible for
pharmacotherapy is informed in an expedient manner while in the
process of patient evaluation. The integration of well-characterized
models that account for sources of variation in PK, PD and/or
relationships with clinical outcomes are integrated into this system
along with the most relevant clinical data associated with a particular
or combination of drug and disease states.
DESIGN ARCHITECTURE
SCM
EDW
P.K.B
Database Tier
(DataMart)
Staging
Area
IDX
Middle Tier
Graphing
Engine
Business Logic
(Alerts, Filters, Derived Values,
Summarizations, Oracle Queries)
Workbench
Predictions
Presentation Tier
Graphical User Interface
(Web presentation; Documents)
Database tier consists of patient records from our
electronic medical records system - Sunrise Clinical
Manager (SCM) extracted, transformed and loaded
into a relational table. Views and queries are created
from the relational tables. A multidimensional data
mart that will permit ad hoc queries and analysis. The
middle tier consists of alerts, filters, summarizations,
derived values and predictions. Predictions are
conducted in an external computational platform
(M&S workbench) that executes code in a variety of
languages provided they can run in a batch mode. The
M&S workbench currently accommodates many
standard prediction engines used to forecast PK and
PK/PD relationships (NONMEM, SAS, SPLUS and
R). Prediction engines are gated in the middle tier
through logic that ensures that minimally required
data sets are available for each patient or sets of
patients for meaningful analysis. The user interface is
web-based (AJAX and web2.0 standards).
A Chair’s Initiative Project at the Children’s Hospital of Philadelphia
METHODS and RESULTS
CURRENT FUNCTIONALITY: The current PKB environment is
comprised of 4 main areas: Compendial information from Lexicomp
on-line for each drug, drug-specific FAQ sheets, individual drug
dashboards and therapy dashboards. Working prototypes for two drug
dashboards (tacrolimus and methotrexate) are available as well as the
Oncology Therapy dashboard. More information regarding the PKB
project history, advisory boards, team members and project status can
be found at http://stokes.chop.edu/programs/cpt/pkb/index.php.
Working versions of the PKB prototypes will be available soon at
http://Pkb.chop.edu.
DRUG DASHBOARD SELCTION: The criteria used to rank and
prioritize the choice of agents considered for drug dashboard
development is based on four properties: drug utilization, medical
need, guidance outcome value and dashboard viability.
Drug
Utilization
Guidance
Outcome
Value
Dashboard
Viability
Medical
Need
Dashboard
Candidates
Figure 1. The intersection of 4 target
categories is quantified into a
8000
multimetric key performance index
(KPI) used to facilitate dashboard
6000
prioritization. Methotrexate and
tacrolimus rank in the top 5% of drugs 4000
on formulary based on KPI score.
Figure 2. Utilization of tacrolimus (top) and methotrexate
(bottom) with therapeutic area at CHOP over the past 6 years.
KPI scores which are low in number
reflect drugs for which the value of
pharmacotherapeutic guidance is high.
Higher scores are consistent with drugs
that are well managed, understood and/or
not extensively utilized in children in
general. The equation for KPI score is
shown below (Barrett, JPPT, 2008):
KPI (2 * Medical Need  Utilization  6 *GuidanceOutcome)
Figure 3. CHOP maintains an on-line version of the Lexicomp pediatric dosing
compendium that caregivers can access over the hospital intranet. It is not
connected to the EMRs. The PKB pulls the electronic data from Lexicomp and
allows the user to structure the information into the desired layout eliminating
the need for excessive scrolling / searching. Compendial data in this and similar
formats will be made available through the individual and therapy dashboards
either automatically (when a specific drug dashboard is selected) or via user
selection (as in the case of a therapy dashboard).
Figure 8. The PKB maintains an interface to hospital accounting
reports from the Sieman’s accounting database to assess drug
utilization trends and allow caregivers and investigators to examine
time-based changes in prescribing across various strata (i.e, age,
therapeutic area, drug class).
Figure 4. The methotrexate dashboard pulls TDM data, dosing histories,
laboratory values, study protocol, and rescue therapy information indexed on
each dosing event from the EMR system and the CHOP enterprise warehouse
into a single view of the most recent dosing event. Relevant fields such as the
hydration schedule, serum creatinine, and leucovorin dosing are tabulated and
plotted with customization available to the user. Views to historical data (within
and across patients) are available as well. Additional functionality allows the
caregiver to forecast the MTX exposure based on the institution’s monitoring
requirements and examine the individual patient’s performance relative to
established nomogram’s pertaining to leucovorin co-administration.
Figure 9. The PKB ensures cross-talk between dashboard elements
(and the underlying source data) via the therapy dashboard concept –
a holistic approach to assessing the composite drug therapy
administered to a patient. Related diagnoses that can be collectively
assigned to an individual patient’s drug therapy are pooled so that
dosing histories related to TDM and relevant patient markers and
outcomes can be easier to view. Access to available therapy
dashboards is maintained from exiting EMR systems and based on the
patient medical record number.
DRUG ADMINISTRATION - IMMUNOSUPPRESANTS
AZATHIOPRINE
CYCLOSPORINE
MYCOPHENOLATE
SIROLIMUS
TACROLIMUS
2000
0
2000
2001
2002
2003
2004
2005
2006
YEAR
DRUG ADMINISTRATION - ONCOLOGY
3000
2000
1000
ASPARAGINASE.L
BLEOMYCIN
BUSULFAN
CARBOPLATIN
CARMUSTIN
CISPLATINUM
CYCLOPHOSPHA
CYTARABINE
DACTINOMYCIN
DAUNORUBICIN
DOXORUBICIN
ETOPOSIDE
FLUOROURACIL
IDARUBICIN
IFOSFAMIDE
LOMUSTINE
MELPHALAN
MERCAPTOPURINE
METHOTREXATE
PEGASPARAGINASE
PR0CARBAZINE
THIOGUANINE
THIOTEPA
VINBLASTINE
VINCRISTINE
Figure 5. MTX disposition is described by a two-compartment
model with first-order elimination. Although MTX clearance
changes over time in patients with renal dysfunction, clearance is
approximated with a simple model defined by two different
clearance distributions for the two populations. The Bayesian
forecasting model utilizes the NONMEM PRIOR subroutine to
incorporate population priors into the model. The clearance
mixture model assigns patients to a population (normal or
impaired clearance) based on the probability of that patient
belonging to either population given their MTX plasma
concentrations. The MTX dashboard has the capability of
updating the priors based on new patient data (the subject of
ongoing operational validation efforts) and projects future MTX
exposure on the leucovorin nomogram normally used to assess
patient status and confirm leucovorin dosing. Retrospective
validation suggests that the MTX dashboard has the potential to
recommend dosing adjustments and rescue therapy sooner than
standard practice.
0
2000
2001
2002
2003
2004
2005
2006
YEAR
Table 1. Alignment of desired functionality and drug attributes for prototype drug dashboards – Methotrexate
and Tacrolimus
Figure 6. The tacrolimus (TAC) dashboard is customized to show patient data
relevant to TAC clinical performance. For example, the organ type, date of
transplant, and donor source are identified patient descriptors. As the typical
time course of TAC administration is quite long relative to other drugs, the time
scale is scrollable and again linked to tabular output. Forecasting drug exposure
is functional though a final population model has not been validated. Ongoing
collaboration with the Hospital for Sick Kids in Toronto is expected to yield a
final model to guide TAC dashboard predictions.
Figure 7. The TAC dashboard also permits views of exposure, dosing and
biomarker histories along with indices for clinical events. By mousing over the
clinical event data the caregiver can see exactly what the recorded event was
relative to dosing history or TAC levels.
Figure 10. The prototype oncology dashboard provides a userdefined snapshot of dosing histories related to specific cancer therapy.
The system screens for interaction potential based on PK attributes
even if published information is unavailable (i.e., based on common
metabolic pathways). Future versions of this tool will incorporate
Simcyp methodology to create virtual pediatric patients from which
drug interaction potential can be simulated and visualized by the
caregiver.
ONGOING EFFORTS and FUTURE DIRECTION
• Integration of the PKB with the new EPIC EMR system is ongoing and
system testing of the prototypes is expected to commence by year’s end.
• Dashboards for vancomycin and vincristine in progress. Antibiotic Therapy
dashboard planned.
• Integration with additional prediction engines (Simcyp, R, SAS, etc) planned
for 2009; emphasis placed on forecasting drug interaction potential.
• An exportable version of the PKB (PKB-Lite) is being configured to
accommodate small community, outpatient settings.
• Field testing of the PKB integration at other large inpatient settings planned
for 2009.
REFERENCES
Zuppa AF, Vijayakumar S, Mondick JT, Pavlo P, Jayaraman B, Patel D, Narayan M, Boneva T, Vijayakumar K, Adamson PC, Barrett JS.
Design and implementation of a web-based hospital drug utilization system. J Clin Pharmacol: 47(9): 1172-1180, 2007.
Barrett JS, Mondick JT, Narayan M, Vijayakumar K, Vijayakumar S. Integration of Modeling and Simulation into Hospital-based Decision
Support Systems Guiding Pediatric Pharmacotherapy. BMC Medical Informatics and Decision Making 8:6, 2008.
Barrett JS, Patel D, Jayaraman B, Narayan M, Zuppa A. Key Performance Indicators for the Assessment of Pediatric Pharmacotherapeutic
Guidance. J Pediatr Pharmacol Ther 13:141-155, 2008.