An Individualized ESA Dosing Regimen for Hemodialysis Patients
To Stabilize Hemoglobin Levels in a Target Range
James Rogers, MS, MBA
Advance Management Group, LLC
4447 Manchester Lane, NW, Rochester, MN 55901
T: 507-289-4680
rogers.james@amgresults.com
Edward J. Gallaher, Ph.D.
Applied System Dynamics, LLC
533 NE 37th Avenue, Hillsboro, Oregon 97214-6307
T: 503-577-7547; F: 503-648-5637
ed@appliedsd.net
Craig Hocum, P.A. -C
Mayo Clinic
200 First Street, SW, Rochester, MN 55905
T: 507-312-4353; F: 507-266-1963
hocum.craig@mayo.edu
David Dingli, M.D., Ph. D.
Mayo Clinic
200 First Street, SW, Rochester, MN 55905
T: 507-284-3417; F: 507-266-4972
dingli.david@mayo.edu
Introduction
We previously reported on a modeling project designed to determine erythropoietic
stimulating agent (ESA) dosing regimens for hemodialysis patients, (Rogers, Gallaher, and
Hocum). There, we describe the stages of model development, its implementation, and the
results obtained.
Here, we revisit the project to describe the model structure, focusing on the empirical
foundations of the formulations and parameters as found in biomedical and clinical literature.
We describe the model behavior, its replication of the clinical issue at hand, and show how the
model was used to develop superior policies. We then discuss how the new policies were
implemented in the host organization, and present the results which accrued over time. After a
discussion of the limitations and possible extensions to the model, we suggest how so-called
“inside the skin” dynamic modeling could shape the future of individualized medicine.
Contextual Background
End stage renal disease (ESRD) is the last stage of chronic kidney disease (CKD). At this stage the
kidneys are functioning below 10 percent of their normal function, and cannot support a
person’s day-to-day life by removing waste and excess water. ESRD is usually the result of
chronic kidney disease (CKD) which causes declining kidney function to the point of complete
failure.
ESRD imposes a significant burden on the US healthcare system. Modalities of renal
replacement therapy are transplantation, peritoneal dialysis, and hemodialysis (HD). In 2014
the US had 410,000 HD patients, 63% of all prevalent ESRD cases, at the end of 2014. USRDS
ADR (2016).
=m om Al] ESRD sees Hemodialysis == Peritoneal dialysis s====Transplant
6004 eet
Number of patients (in thousands)
TI tt
96 97 98 99 CO 01 O2 03 O4 0S 06 07 O8 O89 10 11 12 13 14
Year
Figure 1. Trends in the number of ESRD prevalent cases (in thousands) by modality, in the U.S.
population, 1996-2014
When Medicare eligibility was extended to persons with ESRD in 1972, only about 10,000 were
receiving dialysis. This patient group grew to nearly 400,000 patients by 2014. Even though the
ESRD population has remained at less than 1% of the total Medicare population, it has
accounted for about 7% of Medicare fee-for-service spending in recent years. The cost to
Medicare at year end 2014 was approximately $26B, as shown in figure 2. USRDS ADR (2016).
40
@ Transplant
@ Peritoneal dialysis
@ Hemodialysis
Total expenditures (,in billions)
oO
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Year
Figure 2. Total Medicare ESRD expenditures, by modality, 2004-2014. Cite USRDS.
It is beyond our scope to provide a detailed analysis of the Medicare reimbursement policies for
ESA use for HD patients. Summarizing, the cost of ESA “injectables” for HD patients had grown
to just under $2B by year end 2010. A bundled payment system was implemented in year 2011
after which Medicare reimbursements for these drugs were no longer tracked. For our
purposes we note that ESA expenditures were indeed significant and any improvement in their
use would be of great value.
HD therapy in 2014 was primarily performed by 6,750 in-center Dialysis Care Facilities (DCF)
organized into 18 regional networks. In 2014 the two largest dialysis organizations, Fresenius
and DaVita, collectively treated 69% of patients in 65% of all dialysis units.
HD patients typically receive therapy 3 times per week in a 3- to 4-hour session at their
assigned DCF.
The quality of Hemodialysis therapy is under constant review, with a focus upon these key
indicators:
Dialysis Adequacy: the extent to which removal of wastes and excess fluids is achieved per
dialysis session.
Access Method: how the patients’ bloodstream is accessed and routed to external filtration,
with a preference for a surgically implanted fistula versus an inserted catheter, which is more
susceptible to infection.
Bone Health: Various lab measurements involving calcium, phosphorous, and parathyroid
hormone
Infection Rate: With a goal of zero.
Anemia Management: as measured primarily by a patient’s Hemoglobin (Hgb) conformance to
a target range. At the initiation of this project, the nationally recommended target range for
Hgb was 10-13 grams per deciliter of blood volume.
The Clinical Setting
This project addressed anemia management issues for HD patients receiving care through Mayo
Clinic Dialysis Services (MCDS), the service arm of the Mayo Clinic’s Department of Nephology
and Hypertension.
By way of introduction, MCDS is classified as a Small Dialysis Organization (SDO) by the Centers
for Medicare & Medicaid Services (CMS), having fewer than 100 DCF’s in service. MCDS is part
of:
* Anacademic, non-profit institution, caring for
* 650-700 prevalent HD patients, across
* 17 DCFs, staffed by
* 15 nephrologists, 7 allied staff - 2 PAs, 2 NPs, 3 RNs, following
* Common policies & procedures, supported by a
* — Shared dialysis database
At the initiation of the project in 2008, MCDS anemia management performance goals were:
Best Practice: 85% or more of dialysis patients with mean Hgb [average of first-of-the-month
values for previous three months] between 10 and 13 g/dl, and
Recommended Practice: 80% or more of dialysis patients with mean Hgb between 10 and 13
g/dl
At that time, 34% of MCDS hemodialysis patients had mean Hgb levels above 13, 59% had levels
within the target of 10 to 13, and 7% had levels below 10. (McCarthy.)
Plans were underway within CMS at that time to initiate bundled pay for performance
reimbursement policies in 2011. Reimbursements to providers would be reduced if
performance objectives were not met. It was important to MCDS to understand why anemia
management performance levels were below target and develop strategies to improve them.
A Problem within the Problem
Hgb values are reported in g/dl, while hematocrit denotes the volume percent of packed RBCs;
both measurements refer to whole blood samples. Normal values for healthy adults lie within
the ranges shown below (Billet).
Hgb Hematocrit
Male 14-18 g/dl 40-54%
Female 12-16 g/dl 36-48%
Table 1. Normal Hgb and Hematocrit Ranges
Healthy adults have a total of about 2-3e+13 red blood cells (RBCs), consistent with the Hgb and
hematocrit values shown above. RBCs have a lifespan of about 120 days; senescent cells are
removed by the spleen and iron is recycled for use in new cells. RBCs must therefore be
replaced at a rate of 2e+6/second, or 1.7e+11/day (Elliott).
RBC production takes place in the bone marrow and is critically dependent on the hormone
erythropoietin (EPO). EPO is synthesized and released into the circulation by specialized cells in
the kidney as a response to tissue hypoxia, thus creating an endogenous balancing loop to
maintain adequate tissue oxygenation. Although the pathogenesis of anemia of CKD is
multifactorial, the decreased production of EPO with declining renal mass is considered the
primary etiologic factor.
For reasons that are unclear, optimal Hgb values are somewhat lower for dialysis patients,
between 10 and 12 g/dl. Values below 10 g/dl are evidence of anemia, leading to poor tissue
oxygenation, fatigue, cognitive deficits, dyspnea, an increased likelihood of falls, along with a
worsening quality of life. Hazards of levels above 12 g/dl are less obvious; however, increased
blood viscosity contributes to an increased probability of congestive heart failure, heart attacks,
and strokes.
Recombinant human EPO (rHuEPO; Epogen) has been cloned, and was approved for use in 1989
with the following indication: “treatment of anemia associated with chronic renal failure,
including patients on dialysis (end stage renal disease) and patients not on
dialysis.” (Luksenburg). Along with iron, exogenous replacement of EPO quickly became a
standard of care for ESRD patients. (Kalantar-Zadeh). A derivative of EPO, darbepoetin-alfa
(Aranesp) became available in 2001, with similar indications.
In this paper we refer to these drugs as erythropoietic stimulating agents (ESAs). Millions of
people worldwide have benefited from these drugs over the past 28 years. Before their
availability, the standard of care for the treatment of anemia among dialysis patients was blood
transfusion, a therapy which included risks of immunologic sensitization, infection, and iron
overload.
Results of pre-approval Phase | and II clinical trials conducted in 1987 demonstrated that
recombinant human erythropoietin [ESA] is effective, can eliminate the need for transfusions ...
and can restore the hematocrit [and Hgb] to normal in many patients with the anemia of end-
stage renal disease. (Eschbach et al). Figure 3 shows how effective ESA therapy was — it was
revolutionary.
Treatment of an anemic patient with
Normal | transfusions (Tx) and then with ESA
Red Blood Cell Count
hk i Ik ik ESA Therapy
Jan Feb Mar Apr May Jun ul
Figure 3. ESA replaces transfusions as the preferred standard of care. Adapted from
Eschbach.
Treatment with recombinant human erythropoietin (r HuEPO) [ESA] has been a major advance
for the management of anemia in patients on hemodialysis. Therapy, however, is typically
observed to be associated with recurrent cyclic fluctuations in hemoglobin levels. It is most
closely associated with frequent rHuEPO dose changes, hospitalization, and iron treatment
practices. (Fishbane).
The Clinical Challenge
In addition to the problems with low and high Hgb levels described above, Hgb cycling itself is
detrimental. Blood viscosity is highly sensitive to hematocrit, creating additional burdens on the
cardiovascular system.
Late in 2007, we began an effort to understand why so many MCDS patients had Hgb values
outside the target range through a comprehensive review of the available data. We too
observed Hgb cycling, with many patients having a longitudinal history Hgb lab value
measurements similar to those of figure 4.
A Frequently Observed Example of Hgb Cycling
ESA Dose
672 Days
—@Hgb —Hgb Lower Limit ——Hgb Upper Limit —*—ESA Dose
Figure 4. Typical pattern of Hgb cycling among MCDS HD patients following standard
protocol. (McCarthy).
We had two issues to resolve. We needed to understand why so many patients’ Hgb lab values
were out of range, and we needed to understand why individual patient hemoglobin values
displayed cycling.
Our initial hypothesis was that the system shortcoming to be improved was tightly connected
to the standard ESA dosing protocol, shown in Table 2.
Titration of ESA Dosage: Given IV once weekly (| i dose 300 mcg/week)
Hgb < 10 g/dL Increase by 3 vials
Hgb 10.1-10.5 g/dL Increase by 2 vials
Hgb 10.6-11.4 g/dL Increase by 1 vial
Hgb 11.5-12.5 g/dL No change in dose
Hgb 12.6-13.9 g/dL Decrease by 1 vial
Hgb > 14 g/dL HOLD 2 weeks; then decrease by 2 vials
Available Vial Sizes (mcg) 25, 40, 60, 100, 150, 200, 300
Table2. MCDS Standard ESA Protocol. (McCarthy).
The next section describes a compact model that shows how undesired oscillations in individual
patient Hgb levels are caused as a results of the dosing protocol in place.
A Model that Explains Why Hgb Oscillation Occurs and How to Prevent It
Having observed many examples of hemoglobin cycling, the consulting team, though originally
engaged to develop a performance measurement system, recommended a joint modeling
project to develop a System Dynamics model to explain and eliminate the observed oscillations.
The purpose of the model was to explore the relationship between periodic Hgb cycling and
ESA therapy and identify interventions to dampen the cycling.
Details of the model building process are described in (Rogers, Gallaher, and Hocum). The
model building process led us on an exceptional journey which led to dramatic improvements in
patient care, despite our initial lack of subject matter expertise. The reader is encouraged to
review the incremental steps of model construction.
The model is based upon the process of erythropoiesis — the clinical term for the supply chain
that continuously refreshes the supply of the body’s red blood cells (RBC’s) as they expire.
Hemoglobin is a measure of RBCs in circulation — both mature RBC and newly formed RBCs,
called reticulocytes. Figure 5 portrays the process as found in any standard hematology text
(see Kaushansky et al).
About 3 -4 Days
i dependence
Erythropoietin
@.3 -GP -D-
6
«-—— About 10-13 es
Days
f About 21 Days |
About 1-2 Days
EPO dependence
Pluripotent Burst-Forming Colony-Forming Proerythro- Erythro- _Reticulocytes RBCs
Stem Cell Unit-Erythroid Unit-Erythroid blasts blasts
(BFU-E) Cells (CFU-E) Cells
Figure 5. Erythropoiesis.
The important message in the diagram is that RBCs develop in stages in a process requiring 21-
25 days. Increases in RBCs, as measured by increases in Hgb, resulting from EPO (ESA)
administration will not be observed for more than two weeks. Coupled with the knowledge
that RBC response to ESA administration is not only delayed, but nonlinear in both dose
magnitude and frequency of administration, we have a classic situation leading to overshoot
and undershoot of Hgb levels, a situation highly amenable to System Dynamics modeling.
Through many iterations over a several months we developed a functioning stock and flow
model. Figure 6 provides a conceptual overview.
Bone Marrow Circulation
BFU Production
Replicating Reticulocyte Maturing
CFUs Development Recticulocytes RBCs
‘s
CFU Input Erythroblast Reticylocyte RBC RBC
Production Release Production Lifespan
Hemoglobin
©
evel Exogenous
EPOR Muttiplier Recticulocyte P Medical
Dosing
Orders Factors
Survival
CFU SS
ae
Eliminating
ESA Halt Life
Figure 6. Conceptual Overview of the Model. (Rogers, Gallaher, and Hocum).
The model has two compartments. The Bone Marrow, where erythropoiesis unfolds, includes
all of the progenitor and for precursor cell types indicated in figure 5. The circulating blood
includes reticulocytes and RBCs, giving rise the Hgb measurements.
In contrast to figure 5, the model assumes only the CFU cell type is responsive to ESAs. The
effect of ESAs is to prevent cell apoptosis or programmed cell death. Thus ESAs stimulate the
eventual increase in RBCs by allowing more CFU cells to survive. CFU cell types undergo
proliferation for 12 days in this model, dividing at intervals of 24 hours. The outflow,
Erythroblast Production, contains 212 times are many cells as the inflow CFU Input, less the total
number of cells destroyed by apoptosis, as moderated by ESA levels.
Reticulocyte Development has a duration of 6 days. This stock represents all proerythroblast
and erythroblast cells indicated in figure 5. At this stage several changes to the cell occur, the
most notable being incorporation of iron, “hemoglobinization” and enucleation — the expulsion
of the cell nucleus. (RBCs have no nucleus, which is why they do not replicate but expire at the
end of their lifespan.) If iron bioavailability is not adequate, not all reticulocytes survive.
Maturing Reticulocytes are assumed to develop in RBCs in a 2 day period. RBCs exist in
circulation in healthy patients for about 120 days. The lifespan of RBCs among HD patients is
typically ranges from 60 to 90 days.
Model Parameters
We included five parameters in this model: BFU Production, CFU Survival, EPOR Multiplier,
Reticulocyte Survival, and RBC Lifespan. Each of the model parameters was designed to
correspond to a biophysical process occurring within the erythropoietic process.
To develop an understanding of why model parameters have specified ranges, we explain
below how a truncated Monte Carlo simulation process was used to determine a useful set of
model parameter values. Here, we describe the parameters.
BFU Production. For reasons to be described below was assumed to be some value between
5e+7 and 1e+9 per day.
CFU Survival. This parameter was selected from a range of 20% to 80% of a subpopulation of
the CFU cells. Cells marked for survival develop EPO receptors on the cell surface. If EPO is
present it binds to the EPO receptor and initiates a series of intracellular events that protect the
cell from programmed cell death (apoptosis).
EPOR Multiplier. This parameter relates to an intracellular process which occurs in
erythropoiesis. When EPO binds to an EPO receptor on the cell surface that sets up a cascade
of events within the cell. The effect is to multiply the cell sparing affects by some value
between 1 and 10.
Cell sparing in the model occurs then based upon the fraction of cell surface EPO receptor cells
bound and the magnitude of the subsequent intracellular processes.
Reticulocyte Survival. This parameter was selected from a range of 20% to 80%.
RBC Lifespan. This parameter was assumed to have a range of 40 to 100 days.
As we explain below, a given set of model parameter values constitute a response profile of an
individual to the presence of ESAs. For an individual patient, when each of the parameters is
set to a specific value within its respective range, a simulated specific pattern of RBC production
will occur over time in response to ESA dosing.
Model Boundary
After an exploration of the literature describing the relationship between erythropoietin and
hemoglobin, we chose the boundary of the model to include only a part of the ESA-dependent
stages of RBC development and the successive stages of erythropoiesis which lead to RBC
creation. The ESA dependent stages we chose to include in the model are the stages in which
ESA sensitivity is at a maximum.
There are important exclusions from the model boundary which we list here. The model
assumes:
Iron Sufficiency: The patient to be modeled has adequate iron stores, having a transferrin
saturation lab of 20% or more.
B12 and Folate Sufficiency: the patient to be modeled has adequate B12 and Folate stores.
These affect cell division and replication, inadequate stores lead to compromised RBC
production.
Stable Medical Condition: The patient’s medical condition is relatively stable, currently
unaffected by infection (which sequesters iron making it unavailable for erythropoiesis) and
free from internal bleeding.
Blood Sampling for Hgb Measurement: The hemoglobin blood draw is always obtained at the
conclusion of a hemodialysis session, after excess fluids have been removed, avoiding a
laboratory distortion known as hemodilution.
IV Only: The ESA is administered intravenously, not subcutaneously.
Compliance: The recommended dosing regimen in terms of the amount and frequency of the
drug to be administered will be precisely followed.
Thus, in comparison to the complexity of real world erythropoiesis the model is quite simple,
simulating only the effect of ESAs on RBC creation and assuming all other factors of RBC
production are operating normally. Though the model has limitations, pilot studies indicated it
captured the dynamics with sufficient resolution to be clinically useful.
Referring to figure 6, the model requires the clinician to confirm the dose recommendations.
As we describe below, the scope of factors included in this decision increased over time as we
gained experience in using the model. These decisions involved real-world adjustments to
variables outside the boundary of the model such as iron administration, or interventions to
address detected inflammation or GI bleeding.
How the Model Was Used to Develop Individualized Dosing
To describe the behavior of the model, we next to explain how the model was used in two
distinct phases.
In the descriptive phase we entered historical ESA doses and resulting Hgb levels from a prior
period. We sought a set of parameter values such that known historical ESA doses produced the
known historical Hgb levels. This was accomplished using a Monte Carlo strategy consisting of,
typically, 100 simulations. In each simulation the model generated a set of random initial
parameter values, drawn from a range of values known to be feasible based on current
understanding of erythropoiesis. For each known (historical) Hgb value, the model accumulated
the squared difference between the clinical value and the simulated value. After all the
simulations were complete, the sets of random parameter values and their associated MSE
values were exported to Excel. The data were sorted to find one or more parameter sets that
yielded a minimum Mean Squared Error (MSE) of simulated hemoglobin values in comparison
to the historical hemoglobin values.
The output of the descriptive phase is a set of parameter values which we regarded as an
individual patient’s response profile to the drug. We note that this solution to the matching
problem is not unique. There may be other sets of parameter values which also describe the
patient’s response profile.
In a successive prescriptive phase, we assume that the patient’s response to ESAs will, in the
immediate future, be the same as their response in the immediate past, provided the patient’s
medical condition has not changed.
To execute the prescriptive phase, the patient’s model is initialized with parameter values
identified in the descriptive phase. By trial and error, constant weekly dosing regimens were
simulated to find a weekly dosing regimen which stabilizes the patient’s simulated hemoglobin
at a desired target level. We call this the weekly therapeutic dose (WTD). In most cases, the
WTD so identified was different from any of the available vial sizes.
As shown in table 2, the drug is only available in seven discrete vial sizes: 25, 40, 60, 100, 150,
200, and 300 mcg. Because the WTD we obtained as an output of the prescriptive phase was a
continuous variable, as a final step we titrated the available vial sizes, scheduling a combination
of various vial sizes to be administered at frequencies which would deliver an effective dose
equivalent to the WTD.
As an example, if the WTD determined for an individual patient was 35 mcg per week, we used
the patient-calibrated model to experiment with various combinations of standard vial sizes to
arrive at a recommended prescription. In this example, that recommendation might be
something like “to achieve an equivalent effect of 35 mcg per week, give 25mcg in week one
followed by 40mcg in weeks two and three and repeat three weeks cycle: 25, 40, 40.” It was our
experience that we could always find a combination of standard vial sizes which produce a
simulated projected hemoglobin pattern which identically matched the simulated hemoglobin
pattern over time in response to administration of the any “non-standard” WTD.
In the analysis of the model’s behavior we discuss next, we refer only to the WTD.
Behavior of the Model with Respect to the Standard Protocol and a new Model-Based
Protocol
With this background in mind we can now turn to an exploration of the model’s behavior in
response to the dosing protocol of record.
Consider a patient for whom historical ESA and hemoglobin data has been processed in the
descriptive phase. A set of parameter values, the patient’s response profile, is available to
simulate how the patient will respond to any dosing regimen. In particular, we can simulate
how the patient would be expected to respond to ESA dosing as defined by the standard
protocol described in table 2.
We also can readily determine the patient’s WTD to stabilize Hgb value at any desired target
level.
Using this information we present nine cases to help answer three questions for each case:
- When simulating the standard protocol, does the model replicate the often observed
hemoglobin cycling? If so, under what conditions?
- Does the model allow us to determine a dosing regimen, a so-called “model based protocol”,
which stabilizes hemoglobin at a desired level?
- Does the model allow a quality comparisons between a standard protocol and the model
based protocol?
The nine cases we consider are shown in table 3:
Hgb Hgb Hgb
Falling Stable Rising
CASE 3 CASE 9
Hgb ihe a ee. ——T
Above
Target
CASE 4 CASE 1
Hgb UPPER LIMIT
Within
Target LOWER Limit
CASE7 CASE 2 CASE 6
Hgb
Below
Target co
——— _
Table 3. Nine Cases for Comparison of the Standard Protocol to the Model Based Protocol.
For each case we show the results of two simulations. First (Part A) a simulation of the patient’s
response to the ESA dose determined by the Standard Protocol, and second (Part B) a
simulation of the patient’s response to the WTD.
These nine simulations were performed using the same patient model. Parameter values we
obtained from the descriptive phase for this patient are shown in table 3. The WTD for all cases
except Case 6 and Case 7 had been determined to be 20 mcg per week. The WTD for this
patient for Case 6 and Case 7, using a slightly modified set of parameter values to allow the Hgb
level to drop sufficiently low for testing, had been determined to be 50 mcg per week
Parameter Values for Cases 1-5, 8, 9 Values fir Cases 6 and 7
BFU Input 6.5 x 10° 5.5x 10°
CFU Survival 0.41 0.41
Reticulocyte Survival 0.46 0.46
EPOR Multiplier 2.45 2.45
RBC Life Span 65 65
Table 4. Model Parameter Values Used for Testing Cases 1-9.
The results of simulating the standard protocol as compared to the model based protocol for
cases 1-9 are presented in the appendix.
We summarize the main conclusions of the analysis here:
- The standard protocol more often than not leads to oscillation. The standard protocol
stabilized Hgb values only when Hgb values were already stable.
- The WTD derived from the simulation returns the patient to the precise target range in every
case, after a delay.
- The primary reason Hgb cycling occurs is that the current protocol does not consider the delay
between dose administration and observable results.
- If the clinician is only reviewing Hgb values for a one or two month period, the oscillation is
imperceptible because the period of the oscillation ranges from 100 to 220 days. A behavior
over time chart of sufficient temporal duration is necessary to observe what is happening with
the patient.
The Model’s Scope of Application within MCDS
In the previous section we described the behavior of the model for a specific patient. Our
understanding of the model’s behavior advanced through several stages of development. In
this section we trace a pathway we followed in order to convince ourselves of the model’s
utility in treating anemia among HD patients for the entire MCDS patient population.
We exercised great skepticism in the beginning. We reviewed two or more years of
longitudinal data for each of 50 patients at the DCF attended to by the physician assistant on
the modeling team to assess their compliance with the restrictions we had placed upon the
boundary of the model. We developed detailed BOTs, an effort in and of itself that was quite
revealing. In addition to patterns of Hgb oscillation, the charts also revealed inconsistencies in
iron maintenance therapy and recurrent patterns of hospitalizations. The BOT concept which
the modelers introduced to the clinicians eventually became a critically important clinical
document which has since been institutionalized.
We selected 12 patients from the group of 50 as candidates for modeling.
Three members of the modeling team each modeled the 12 patients and compared results.
The three sets of recommendations turned out to be quite consistent. Based upon these, the
authorized physician assistant selected a recommended dosing regimen. Weekly hemoglobin
measurements were taken over a 12 week period to confirm the response matched the
projections of the model. We found that the actual hemoglobin levels matched the projections
of the model.
Figure 7 illustrates a typical outcome we experienced in applying the model based protocol, not
just for the 12 patients described above but for all of the patients we eventually came to
include in the scope of application throughout MCDS.
Typical Response to (Almost) Following the New Protocol
147 7 300
Descriptive Phase Prescriptive Phase
a 4 250
12 +
= ed wee eee ee = 200 @
a1 f+ 8
2
» 7 150 0
10 be
a
V | 100
of
a 7 50
7 Lo FA a a
Matching descriptive phase 360 Days
simulated Hgb with Minimal MSE
—e—Hgb ——Hgb Hi Hgb Lo — — Projected Hgb —G—Rec ESA —*— Actual ESA
Figure7. Example of Patient Response to Recommended ESA, Including Dose Misadministration.
We expanded the initial set of patients to include most of the initially reviewed 50 patients.
(While we were testing the model for the initial 12 patients, the physician assistant of the
modeling team had addressed some of the exclusions we had observed earlier, the primary
issue being inconsistent iron maintenance.) The majority of the 50 patients were now eligible
for modeling, the main exclusions being either severely compromised medical conditions, or
lack of available data due to recent admission to hemodialysis therapy.
We modeled these 50 patients and begin to outline a process for model based anemia
management: secure and cleanse the most recent data, identify the patient response profile in
the descriptive phase, identify the patient’s WTD and equivalent titrations of available vial sizes,
review dosing recommendations with the prescribing authority, order and administer the
prescription, capture weekly data to confirm adherence to the recommended dosing regimen
and Hgb conformance to the model’s projection, review and discuss among the modeling team,
and identify any required corrective actions.
16
This process generated requirements for what would eventually become a web-based
application to manage all the information accruing for the scope of patients included. The
process as we followed also became the coaching and education template we would use in the
rollout the anemia management protocol to other MCDS DCFs.
We reported our early successes to the division chair, requesting that we expand the scope to
include more patients from other DCFs within MCDS. In response, the division chair requested
that we first identify “patients for whom the model did not work” and provide the reasons why.
We did find 18 patients for whom actual Hgb in the observation period did not match the
model's projections. We found that 13 of the 18 patients had dosing variances from the model
based dosing recommendations (similar to the example in Figure 7) and four of the patients had
emergent medical conditions. The reasons for the one remaining patient’s variance from the
projection remained a mystery at the time. Based upon these confidence-building results as to
the usefulness of the model, the division chair approved our request for expansion.
We then began a process of including all of the patients at all of the other DCFs with the MCDS
system — approximately 650 patients at 15 different locations. The vast majority of the
required data resided in a central database. There were however some exceptions which
required a weekly manual creation and management of some missing data.
We enrolled all patients in the system over a six month period. Between 2009 and 2015, when
the modeling team transferred operational support to Mayo Clinic, over 1000 patients had
been processed using the original model without modification. The purpose of the model was
to develop individualized ESA prescriptions for hemodialysis patients. The structure of the
model allowed us to extend its application to all patients in the system.
A New and Different Policy for the A i for H dialysis Patients at MCDS
Insights from the modeling process together with the proven results which have accrued over
time have brought about significant policy changes in both perspectives and practice. In this
section we describe changes to the anemia management policy and note significant differences
from the past.
In the following section we will discuss how policy resistance was addressed.
The modeling process revealed a powerful insight which in hindsight it is startlingly obvious: for
medically stable patients, a stable and consistent ESA dose produces a stable and consistent
Hgb response. A classic System Dynamics AHA!
This fact is not always been obvious. Fishbane (2005) on Hgb Cycling reports “the cause of
hemoglobin cycling appears to be multifactorial. We found changes in rHuEPO [ESA] dose to be
the most important driver, associated with hemoglobin excursions in approximately 80% of
cases.” In the past, dosing adjustments were made based upon concurrent Hgb conformance
to a target range, apparently unaware of how delayed feedback caused additional and
unnecessary dosing adjustments, which drove cycling.
In the new policy, identification and maintenance of a stable dose is the priority. Adjustment of
the ESA dosing level in less than 4 to 6 weeks is strongly discouraged.
The new policy is informed by the effect of inherent delays in erythropoiesis. It recognizes that
the time between dose administration and observance of the resulting RBCs can be up to three
weeks in duration. In addition to the implications this has for scheduling dosing adjustments,
this delay also implies that historical data in the form of a BOT is essential to understanding the
patient’s condition. In the past, clinical information systems presented only current data, with
historical data available as an option. The new policy requires that historical data be presented
up front. We recommend maintaining a customizable BOT chart spanning at least two years as
a standard clinical document.
In the past, the phrase “EPO resistance” was used to describe patients who seemed not to
respond to ESA therapy. The modeling process revealed that even if adequate ESA doses are
administered, iron deficiency will interfere with erythropoiesis. If the timing of iron
bioavailability in erythropoiesis is not understood, the clinician may erroneously attribute lower
than expected hemoglobin values to EPO resistance. We have observed situations in which the
ESA dosing levels were increased to overcome “EPO resistance” when in fact the issue was iron
deficiency. The new policy insists that adequate iron stores be maintained at all times.
Similar requirements apply to bioavailability of vitamin B12 and folate. Vitamin B12 and folate
are required for proper cell division upstream of the proliferation of the CFU colonies.
Deficiencies can also render ESA therapy apparently ineffective. Macrocytic anemia produces
red blood cells which are abnormally large. This results from improper cell division early in the
erythropoietic process, which can occur with vitamin B12 and folate deficiencies.
Therefore, the new policy requires that before administering ESA therapy, the clinician must
ensure that the patient is iron replete, free from infection, has adequate B12 and folate stores,
and is not suffering from internal bleeding.
Hemodialysis quality management reporting agencies require quarterly reports of the
percentage of patients who are iron replete. Examination of the BOTs for some patients
seemed to indicate that iron deficiency was discovered only after obtaining the required
quarterly lab measurements, as evidenced by quarterly courses of iron supplements once the
deficiency was observed. In the new policy, iron maintenance is recognized as a continuous
process aimed at maintaining a constant supply of iron for erythropoiesis.
After a patient’s Hgb level has stabilized, the new policy recommends careful observation of
small variations in Hgb levels. These are signals to potentially changing medical conditions that
were previously lost in the noise of Hgb oscillation - signals reporting the condition of the
patient. A decrease from 11.5 to 11.2 over one week’s time, for example, could be a signal that
18
the patient is developing infection or there might be bleeding internally. In the past, these
signals were imperceptible.
The goal of the new policy is to do much more than understand what dosing is required to
ensure a patient’s Hgb level is within a desired target range. The goal of the new policy is to
understand what is happening with the patient. The new policy allows for improved quality of
care at a lower cost.
a
A ia Mar Policy i e and R
In hindsight we realize there was little resistance to implementation. The modeling project was
undertaken in 1Q08 and complete by the end of 2Q08. By the end of 4Q08, two pilot studies
had been conducted, a detailed start to finish modeling process was in play, and a web-based
information management application was up and running. Organization-wide enrollment of 650
patients was complete by the end of the second quarter of 2009. Though it did not seem like it
at that time, we now realize that actually was a rapid deployment of a radically different
process!
This is not to say that there were large numbers of people involved in the deployment — there
weren’t. There were several constraining factors which could have extinguished the project,
but we enjoyed having several critical assets which served to overcome them all. First was the
enthusiasm and commitment of the modeling team which was continually reinforced by the
power of the System Dynamics modeling process, the second critical asset. Third, we had open
minded subject matter expertise available to inform model construction, coupled with the
fourth asset, data availability combined with expert data management and programming skills.
Finally, the supervisory environment encouraged our pursuit of the model based process
improvement project with zero micromanagement.
The mild resistance we did experience is common to the obstacles experienced in the
implementation of any new process within the operations of a complex organization. For
example: practitioners were not accustomed to interpreting BOT charts; awareness of the
dynamics generated by delayed nonlinear feedback needed to be taught and nurtured; some
recommended prescriptions seemed misguided - why decrease the dose of a patient with Hgb
within the target range, but for another patient with Hgb also in the target range increase their
dose? We used the model and the tracking tools we had implemented to explain the rationale
of what at first seemed counterintuitive to the clinical team.
There were a few prescribing practitioners who continued to use the then current protocol, not
because they were actively resisting the new one, more likely they were simply unaware. Since
the patients under their care were also enrolled in the system we were to track their data and
construct their BOT charts. Over a period of a few months the typical oscillations associated
with the current protocol presented themselves in contrast to other patients who presented
stable Hgb values. The nursing staff who supported these practitioners were was extremely
helpful in building awareness and encouraging adoption of the new protocol.
By the end of 2Q10, the new anemia management protocol was in play throughout the
organization.
We used three main avenues to overcome the obstacles to implementation: active executive
sponsorship, operational facilitation, and active education and coaching.
Active Executive Sponsorship. Once his confidence in the model was firmly established, the
division chair became quite engaged in the project. He took extraordinary measures to
champion the project among the entire medical direction staff of the Mayo Clinic Department
Nephrology. With only minimal support from the modeling team, he developed and delivered
an hour-long “grand rounds” presentation which fully described not only the benefits of the
emerging anemia management protocol, but also the System Dynamics of why it worked. In
that presentation, he clearly explained how RBC supply chain management is the same process
used by grocery stores to maintain adequate stocks of Rice Krispies on their shelves!
Following that, in November 2010, he presented the “Mayo Clinic Anemia Management
System” publicly at a National American Society of Nephrology conference to an audience of
over 100 practitioners, a presentation we have frequently referenced here.
Operational Facilitation. The data management procedures for collecting, cleansing,
generating, analyzing, and reporting were peculiar to Systems Thinking and System Dynamics.
To get the new system up and running, the modeling team chose to design, implement, and
operate the new information management system as part of the project rather than tossing it
over the wall to the I/T department. This created a rapid application development scenario in
which successive iterations of the information system could evolve without delay.
Active Education Coaching. The three person modeling team developed a weekly schedule of
site visits and web meetings to provide education for the patient facing clinical support staff. A
typical agenda for these meetings consisted of tracking the progress of previously modeled
patients, reviewing new recommendations and their rationale, discussing exceptional patient
situations, and designing corrective actions and interventions. The cumulative effect over time
was to create a staff proficient in using the new anemia management protocol who, in turn
provide support and coaching to medical directors.
Cost Reductions from the New Policy
Recall that at the beginning of the project, with respect to the target range for Hgb of 10-13,
34% of MCDS hemodialysis patients had mean Hgb levels above 13, 59% had levels within the
target of 10 to 13, and 7% had levels below 10. After rollout of the new process in 2Q09, only
11% of patients had Hgb levels above 13, 80% had levels within the target range and 9% had
levels below 10.0 (McCarthy.)
20
This was a significant improvement which continued over the successive years.
Though cost reduction was not an objective of the study, we did realize significant reductions in
anemia management costs. Referring back to the available vial sizes in table one, the use of the
three largest vial sizes was all but curtailed, not because it was the goal but because those dose
levels were no longer required to stabilize Hgb values.
ESA and Iron Costs per Patient per Month
$800 - 738
$700 +
$600 - 666
$500 + 455
$400 +
$300 F 362
$200 + » ag
$100 } — |
so 1 1 rm i i J
1HO7 2HO7 1HO8 2HO8 1HO9 2HO9
—e ESA —elron —eTotal
Figure 8. Reduced Cost of Anemia Management Injectables After Implementation of
the Model Based Protocol in 2008. (McCarthy)
Anet reduction of $283 per patient per month in anemia management drug costs was observed
for 350 MCDS patients, producing an annual savings of $1.3M per year. If we extrapolate this
figure to the full set of 650 patients we were serving, savings amount to as much as $2.4M
million per year.
Limitations of the Current Model and Possible Extensions to Improve Its Usefulness
The model we have described has many limitations but has proven to be quite useful for almost
eight years of clinical application. In this section we describe how an improved model
addressing some of those limitations might lead to even better outcomes.
Administrative and Outcome Tracking Improvements. The model-supported process
improvement project we initiated in 2008 was focused upon a single purpose: to explore the
relationship between periodic Hgb cycling and ESA therapy and identify interventions to
dampen the cycling.
We did not anticipate how valuable it is to achieve and sustain individualized Hgb levels in
terms of overall patient care. Stable Hgb levels opened a new window into the patient’s
21
underlying condition. We did not realize that this improved care would lead to significant
reductions in hospitalizations as reported in (Rogers, Gallaher, and Hocum). We did not
anticipate the extent to which improved care would most likely improve patient well-being and
survival. Finally, we did not realize that clinical staff productivity concerning anemia
management tasks would improve by orders of magnitude.
Each of those improvements is a direct outcome of applying Systems Thinking and System
Dynamics to this problem, but we did not prepare to measure those improvements,
unexpected as they were.
An improved process deployment project would measure initial baselines for each of those
measures: patient well-being, patient survival, hospitalization rates, staff productivity, ESA
therapy effectiveness, and overall ESA costs (including order rates and safety stocks). Data
collection routines should be updated to include tracing those measures over time. The model
should be updated to include cost and productivity metrics.
Patient Specific Biometrics. As noted above, the window into the patient’s well-being created by
the ability to maintain stable Hgb levels enabled significant improvements in the quality of care.
Occult bleeding was exposed, infections were detected early, and other changes in the patient’s
medical condition became more visible to the clinician. Improvements to the model which
increase its sensitivity to minor changes in Hgb levels might improve its usefulness in
monitoring the patient’s underlying medical condition. One way to accomplish this might be to
incorporate patient specific data rather than using “standard patient” values. For example, in
the current model, the estimated volume of distribution for the ESA is fixed at 7% of total body
weight, which is assumed to be 70 kg. Sensitivity analysis with respect to patient specific
biometrics remains to be performed.
Subcutaneous Administration. Many patients currently receive ESA doses subcutaneously but
the current model only considers intravenous administration. Subcutaneous administration
involves differences in absorption, distribution, metabolism, and elimination which in turn
affects drug concentrations. Different concentrations of the drug change the survival rates of
CFU cells which eventually affects RBC counts and Hgb levels. Inclusion of subcutaneous
administration would also be helpful in model confidence building among clinicians who use
and prefer that route.
Administration of IV Iron. Bailie (2015) writes “Intravenous (IV) iron is required for optimal
management of the anemia in the majority of hemodialysis (HD) patients. While the IV iron
prescription has increased over time, the best dosing strategy is unknown and any effect of IV
iron on survival is unclear ... [studies show] ... increased risks of hospitalization and mortality
with increased dosages of IV iron...In light of these associations, a well-powered clinical trial to
evaluate the safety of different IV iron-dosing strategies in HD patients is urgently needed.
(Italics mine)”
22
The proper administration of iron whether in supplemental or intravenous form has been a
controversial and unresolved topic of discussion for more than 10 years. This model needs to
be extended to include iron dynamics.
We are seeking sponsorship for a project which would make these model extensions to create a
simulation model that would be useful to conducting such a well-powered clinical trial to
identify safe and effective iron management policies.
Endogenous Erythropoietin and Extensions to Chronic Kidney Disease (CKD). The current model
assumes that the residual kidney function of ESRD precludes the production of endogenous
erythropoietin. In fact, ESRD patients with polycystic kidney disease do produce endogenous
erythropoietin. This did not present a problem, the current model simply determined the level
of “complementary ESA” which would be required to stabilize Hgb at a target level.
If however the current model was extended to include the dynamics of endogenous
erythropoietin, it might be of use in monitoring the rate of kidney function decline in stages of
CKD preceding ESRD. By monitoring the rate at which “complementary ESA” must be increased
over time to maintain a constant Hgb level, we might gain insights into observing and slowing
the rate of kidney function decline.
Variable Cell Replication Times and Progenitor Cell Self-renewal. Dingli, et.al. (2007) and Demin
et.al. (2010) propose models of erythropoiesis which differ significantly from the current model.
They explore the consequences of increasingly shorter cell replication rates among increasingly
mature RBC progenitor cells, and the fact that BFU cells have the ability to self-renew. The
current model assumes a cell replication rate of one day for all cell types in the erythropoietic
supply chain and no cells have the ability to self-renew. Incorporating these the dynamics
would allow for more accurate estimates of the number of mitotic events that occur from stem
cells to red blood cells, the rate of replication of each cell type, and the numbers of cell types
present in each stage of development in steady sate conditions among other descriptive facts.
The goal of updating the model this way would not be to get better at modeling the world. The
goal would be to expose new and different parts of the world that we desperately need to
understand. Cancer is uncontrolled cell replication. Improvements to the current model which
incorporate these ideas could improve the usefulness of the model in understanding how to
interact with the dynamics of cell division and proliferation in healthy and disease states.
Other Improvements. We have shown both in concept and application that even simple
biophysical system dynamics models can be of great value in improving the quality of
healthcare and reducing costs. Clinicians who have a systems physiology perspective are far
more qualified than the authors to conceive of new and powerful applications. It is our hope
that the presentation of this material captures the imagination of clinical practitioners in search
of improved protocols to treat issues arising from complex biophysical dynamics such as those
we have discussed here.
23
Discussion
ESA therapy for the treatment of anemia among hemodialysis patients is common. This paper
has presented Hgb cycling as an important problem associated with ESA therapy. We described
the structure and behavior of a model to identify the causes of hemoglobin cycling, a model
which also can be used to identify dosing regimens which eliminate cycling.
An unintended and highly beneficial consequence of stabilizing hemoglobin values involves the
ability to interpret small changes in individual patient Hgb levels, variations previously lost in
the noise of Hgb cycling, as indicators into the patient’s well-being. These indicators allow for
early detection of emergent issues and improved quality of care. We have also shown that the
new approach has also produced significant reductions in costs of anemia management
injectables.
We have described how a new process for anemia management was implemented in a complex
organization, tools and techniques we used to overcome implementation barriers, and
suggested several improvements which could be made to the model to improve future
applications. Some of the applications we identified implied extensions to other fields of
clinical care, such as cancer.
We have also identified several improvements that could be made to the current model which
would improve its usefulness in a variety of settings.
We offer these observations in hindsight concerning the power the System Dynamics
methodology demonstrated to guide this project:
e The standard protocol dosing regimen was based upon an individual patient’s two most
recent Hgb levels which had become the clinician’s range of reference. Actual Hgb
oscillations with periods ranging from 100 to 200 days were imperceptible through such a
narrow window. The simple task of drawing an individual’s BOT clearly revealed the nature
of the problem the model needed to address. The first BOT’s produced created many AHA
moments for clinicians.
e Though the problem surfaced in the nephrology department, the necessary insights for its
solution resided in the hematology department. Crossing the interdepartmental boundary
and engaging hematology subject matter experts very quickly revealed the dynamics of the
underlying process and key model parameters that were required.
e The client members of the modeling team had no previous experience with System
Dynamics. An introduction to stock and flow thinking, including simple examples of how
stock and flow structures give rise to complex dynamic behavior patterns enabled them to
productively participate in, or even occasionally lead, the model building process.
e One client team member has remarked that System Dynamics has “changed his perspective
about everything”.
24
e The modelers had little previous experience with either nephrology or hematology. The
successive steps in the modeling process, such as in Sterman (2000), were sufficient to raise
the right questions and produce the right answers to equip the modelers with the required
subject matter expertise.
e The modeling team included a pharmacokinetic and pharmacodynamic modeling subject
matter expert who brought a keen awareness of the scope of issues to consider with
respect to “what the drug does to the body and what the body does to the drug”. His
insights were critical to establishing a laser focus on the purpose of the model as well as
defining a highly functional model boundary.
e The sequence of stock and flow structures which evolved during the course of the project
were powerful communication tools, enabling the joint modeling team to effectively and
efficiently address critical assumptions and avoid falling victims to common misperceptions
about ESAs.
e The modeling process guided us in the creation of a model which matched its purpose of
both explaining and damping Hgb oscillations. The model became the engine to feed an
information management system reporting both individualized and rolled up performance
information, variances from targeted performance, and corrective action plans. This system
enabled remarkable improvements in staff productivity with respect to anemia
management tasks.
e Atthe conclusion of the process the modeling team had acquired deep operational
understanding of the dynamics of erythropoiesis and the dosing regimens to achieve stable
Hgb levels in a desired target range. This understanding became a critical resource which
facilitated organization-wide implementation to achieve significant performance
improvements for MCDS.
Future Work
We believe that this application of biophysical system dynamics merely scratches the surface of
the potential applications which must be pursued in the future. After presenting this material
at the 29 International Conference of the System Dynamics Society in 2011, we convened a
group of system dynamics practitioners to ask for their perspective of what other applications
would be possible. Within one hour’s time they produced a list of over a dozen potential
application areas including 50 topics. The major areas are listed in Table 4.
25
Application Areas suggested by SD Practitioners
Immunodynamics Body fluid electrolytes
Drug delivery systems Chronic disease progression
Pain management Cancers/autoimmune
Individual psychiatric interventions | Alternative and complimentary medicines
Antimicrobial resistance Receptor dynamics
Physiology of sleep disorders Addiction physiology
Stress response Body fluid electrolytes
Table 4. Potential Inside- the-Skin Dynamic Modeling Areas of Application
Exploring any of these areas would be more than an academic pursuit. Each involves large
numbers of real patients facing real difficulties being treated by real clinicians who are often
following protocols of care based on poorly understood physiological phenomena.
Following disciplined principles of System Dynamics applied to inside the skin dynamics
presents the opportunity to understand and implement real solutions.
Our future work envisions developing a management infrastructure that can rationalize, fund,
and execute a well-designed portfolio of modeling projects to bring the underlying dynamics of
areas like these to the light of day.
We hope that our ESRD example has begun to persuade the reader to consider the potential
System Dynamics has to revolutionize all of medicine in areas like these and others.
26
References
Bailie, G., Larkina, M., Goodkin, D., Li, Y., Pisoni, R., Bieber, B., Mason, M., Tong, L., Locatelli, F.,
Marshall, M., Inaba, M., Robinson, B. Data from the Dialysis Outcomes and Practice Patterns
Study validate an association between high intravenous iron doses and mortality. Kidney
International (2015) 87, 162-168.
Billett HH. Hemoglobin and Hematocrit. In: Walker HK, Hall WD, Hurst JW, editors. Clinical
Methods: The History, Physical, and Laboratory Examinations. 3rd edition. Boston:
Butterworths; 1990. Chapter 151. Available from:
https://www.ncbi.nlm.nih.gov/books/NBK259/.
Demin, |., Crauste, F., Gandrillon, O., Volpert, V. 2010. A multi-scale model of erythropoiesis.
Journal of Biological Dynamics, 4:1, 59-70, DOI: 10.1080/17513750902777642
Dingli, D., Traulsen, A., and Pacheco, J.M., 2007. Compartmental architecture and dynamics of
hematopoiesis. PLoS ONE 2, e345.
Elliot, S.G., Foote, M., Molineux, G. (editors) Erythropoietin, Erythropoietic Factors, and
Erythropoiesis; Molecular, Cellular, Preclinical, and Clinical Biology, 2"4 edition. Birkhauser
Publishing, Basel, 2009.
Eschbach, J., Egrie, J., Downing M., Browne, J., and Adamson, J. Correction of the Anemia of
End-stage Renal Disease with Recombinant Human Erythropoietin: Results of a Combined
Phase | and II Two Clinical Trial. N Engl J Med 1987; 316:73-8.
Fishbane, S., and Berns, J.S., 2005. Hemoglobin cycling in hemodialysis patients treated with
recombinant human erythropoietin. Kidney Int 68, 1337-43.
Gross, H., and Lodish, H. Cellular Trafficking and Degradation of Erythropoietin and Novel
Erythropoiesis Simulating Protein (NESP). J. Biol. Chem. 281: 2024-2032, 2006.
Kalantar-Zadeh, Kamyar. History of Erythropoiesis-Stimulating Agents, the Development of
Biosimilars, and the Future of Anemia Treatment in Nephrology. Am J Nephrol 2017; 45:235-
247.
Kaushansky, K., Lichtman, M., Prchal, J., Levi M., Press, O., Burns, L., Caligiuri, M., 2016. Williams
Hematology, 9'" Edition. McGraw-Hill, New York.
Luksenburg, Harvey. Safety Concerns Associated with Aranesp (darbepoetin alfa) Amgen, Inc.
and Procrit (epoetin alfa) Ortho Biotech, L.P., for the Treatment of Anemia Associated with
Cancer Chemotherapy. FDA Briefing Document May 4, 2004.
McCarthy, J. The Mayo Clinic Anemia Management System (MCAMS) Improves Achievement of
Target Hemoglobin (Hgb) Values and Decreases High Hgb Values in Chronic Hemodialysis (CHD)
Patients. 2010. Presentation. The American Society of Nephrology Conference. Denver, CO.
27
Rogers, J., Gallaher, E., and Hocum, C. Individualized Medicine and Biophysical System
Dynamics: An Example from Clinical Practice in End Stage Renal Disease. 2011. Presentation.
The 29** International Conference of the System Dynamics Society. Washington, DC.
Sterman, J.D., 2002. Business Dynamics: Systems Thinking and Modeling for a Complex World.
McGraw-Hill, New York.
United States Renal Data System. 2016 USRDS annual data report: Epidemiology of kidney
disease in the United States. National Institutes of Health, National Institute of Diabetes and
Digestive and Kidney Diseases, Bethesda, MD, 2016.
Yang, W., Israni, R.K. et al., Hemoglobin Variability and Mortality in ESRD. J Amer Soc.
Nephrology 18: 3164-3170, 2007.
28
