Ann Oper Res (2015) 230:87–104

DOI 10.1007/s10479-013-1502-5

Passive cold devices for vaccine supply chains

Sheng-I Chen · Bryan A. Norman · Jayant Rajgopal ·

Bruce Y. Lee

Published online: 20 December 2013

© Springer Science+Business Media New York 2013

Abstract A major priority in many low and middle income countries is the eradication of
infectious diseases through an effective vaccination program. However, the lack of reliable
infrastructure, equipment and support often prove to be significant barriers in such countries.
In particular, unreliable electric power and refrigeration equipment are major concerns and
this has prompted research into the development of passive cold storage devices (PCDs)
that are not dependent on electric power and can maintain cold temperatures for extended
periods of time using a passive medium such as ice. The development of PCDs is in its
early stages, and there are many open questions including how they should be sized, the
interaction between price and device size, and how often they should be replenished. So far,
these types of questions have not been systematically analyzed anywhere in the research
literature, and this paper describes a model to address these issues in order to better guide
the development of PCDs in the near future. This paper examines actual use cases with real
data from a representative country, and presents a model that captures the various tradeoffs
that must be considered in developing a suitable PCD while providing guidance on what
would constitute a useful, robust and economical design.

Keywords Vaccines · Supply chain · Cold chain · Cold storage device · Device design

S.-I. Chen
Nestlé USA, 2121 N. California Boulevard, Walnut Creek, CA 94596, USA

B
B.A. Norman · J. Rajgopal ( )
Department of Industrial Engineering, Swanson School of Engineering, University of Pittsburgh,
Pittsburgh, PA 15261, USA
e-mail: rajgopal@pitt.edu

B.Y. Lee
Department of International Health, Bloomberg School of Public Health, Johns Hopkins University,
Baltimore, MD 21205, USA
88 Ann Oper Res (2015) 230:87–104

1 Introduction and background

The prevalence of infectious diseases has long been an issue of grave concern in many de-
veloping countries around the world. One of the best ways to prevent disease transmission is
by implementing effective vaccination programs, and this has long been a focus of govern-
ments as well as international institutions such as the World Health Organization (WHO).
The Expanded Program on Immunization (EPI) was launched by the WHO in 1974 to ensure
that children throughout the world were immunized against the most common infectious dis-
eases (Jamison et al. 2006), and the Global Alliance for Vaccines and Immunization (GAVI)
was created in 1999 specifically to extend the EPI program to the poorest countries in the
world. Despite these efforts there are still millions of children who do not get the com-
plete set of routine immunizations recommended for their first year of life (WHO-1 2009;
Laxminarayan and Ganguly 2011). Indeed, a recent study (Stack et al. 2011) indicates that
increased rates of vaccination across the world’s poorest countries could save 6.4 million
lives and avert 624 million cases of illness over the next decade.
A major challenge facing policy makers and health administrators in developing coun-
tries is that their health care resources are often very limited. The vaccine supply chains
in such places are often severely strained, with shortages of resources such as cold storage
devices, means of transportation, and adequately trained personnel (Kaufmann et al. 2011).
There also tend to be problems with the underlying infrastructure including electric power,
road networks and equipment. There are numerous ongoing efforts to help governments
and other in-country organizations address these issues and help maximize the number of
children in these countries who are fully immunized.
In this paper we address an issue that is critical in many places. Many vaccines consist of
proteins that may rapidly break down and become ineffective when exposed to temperatures
above 10 °C and therefore must remain in environments with strictly controlled temperatures
(WHO-2 2006). Unfortunately, in low and middle-income countries power sources at many
remote locations where vaccines are administered can be quite unreliable, and coupled with
a shortage of refrigerators this can cause major problems. A relatively recent technological
response to this issue has been the push to develop so-called passive cold devices (PCD).
These devices are designed to use superior insulation and a rechargeable cooling medium
in order to maintain low temperatures for relatively long time periods. Their most desirable
characteristic is that they do not rely on energy sources, and they are also easy to rapidly
install at any location. However, since these devices are still largely in development there
are many questions on how they should be sized, where and how in the vaccine supply chain
they should be deployed, and what price point would make them economically viable.

1.1 Cold chain equipment

The choice of cold chain equipment is affected by many factors, such as facility locations,
forecasted vaccine demand, ordering policies, and required storage and transportation ca-
pacities. We begin with a brief review of both conventional cold storage devices and new
technologies. To organize the discussion, we classify cold chain equipment as active or pas-
sive and present various plausible use cases for each type. It should be noted that there
are more different types of existing equipment than those discussed here, we only select
representative examples from the WHO’s list of Performance, Quality and Safety (PQS)
prequalified devices and equipment (WHO-3 2012).
Cold rooms, refrigerators and freezers are examples of active cold devices. These de-
vices rely on power sources and are installed within facilities. A cold room can hold a very
Ann Oper Res (2015) 230:87–104 89

high volume of vaccines, with sizes ranging from 10 to 50 cubic meters in order to take
advantage of economies of scale. In general, most cold rooms are placed at higher levels
in the distribution chain. Conventional vaccine refrigerators and freezers operate by using
electricity, kerosene or propane. Some models use dual energy sources (fuel and electricity)
within the same equipment, and have standby units with automatic start up when the main
energy source fails. For example, the Sibir 170KE model runs on kerosene or electricity. An
ice lined refrigerator (ILR) is another example of a device that provides stable temperatures
when power cuts are a chronic problem. The backup system can provide a temperature range
of 2 °C to 8 °C for up to 24 hours during power shortfalls or outages.
A solar refrigerator can be operated in areas that are off the electricity grid. In general,
the complete system includes solar panels, batteries, and refrigerators. To save on system
costs, battery-free solar refrigerators eliminate the need for batteries and charge controllers.
These systems are highly efficient with low energy consumption but need to be installed by
a qualified installation company to ensure proper functioning of all components. Moreover,
to ensure reliability, a strong routine maintenance program is also a necessity. This involves
regular cleaning of the solar panels, checking cables and connections, and replacing parts
and/or batteries periodically. Additionally, in many of these settings there are few, if any,
maintenance personnel that are properly trained to work on solar equipment. The costs as-
sociated with system initialization and maintenance are also concerns when considering
solar capacity investment.
The availability of reliable energy at a facility can limit the installation of active cold
chain equipment. Many health posts in low and middle income countries have unreliable
electricity, frequent power interruptions or high electricity costs. The term PCD refers in
general, to devices that are designed to be kept cool by using a cooling medium that is a
phase changing material (the most common one is ice) and typically consist of an insulated
container filled with ice packs to keep vaccines cold without using power. There are a va-
riety of passive cold devices available, including vaccine carriers, cold boxes and insulated
containers. A vaccine carrier has a relatively small vaccine storage volume and is usually
designed for transporting vaccines to health posts. A typical vaccine carrier might hold three
liters of vaccines and maintain proper temperatures for up to a few hours. A cold box refers
to a relatively large container (often twenty to thirty liters of capacity) and is typically used
for longer distance transportation. These are better insulated than vaccine carriers and typ-
ically have a number of ice packs located within the cold box. The cold life can be up to
several days depending on usage and the frequency with which the cold box is opened and
closed.
A more recent innovation has been the development of PCDs that use superior insula-
tion and are designed to maintain low temperatures for much longer periods than vaccine
carriers or cold boxes. In a typical scenario, the cooling medium and vaccines are filled in
PCDs at a recharging point, and health workers then move the refilled PCDs along with
vaccination devices and supplies (i.e., syringes, diluents, etc.) to health posts, clinics and
other remote vaccination points. There are different PCD designs that give rise to different
features. Three principal features are required storage duration, vaccine storage volume, and
required cooling medium volume. Lighter, easily portable PCDs are preferred for outreach
or community-based vaccinations, while those stored at fixed locations could be somewhat
bulkier.
An immediate goal is for PCDs to be incorporated into the cold chain as replacements for
stationary cold chain equipment at health posts, many of which are operating in remote ar-
eas. For these facilities, PCDs could be used to ensure the storage and transportation needs
of vaccines at controlled temperatures. Given that most health posts in current EPI vac-
cine networks typically get replenished about once every month, an initial design criterion
90 Ann Oper Res (2015) 230:87–104

in developing PCDs is that they be capable of storing vaccines for several weeks between
recharging. There are several benefits of using passive cold devices for storage and trans-
portation including:
• Ease of access to outreach areas
• Independence from power outages
• Low probability of equipment breakdown
• Low recurring costs (i.e., energy and maintenance cost)
• No need for vaccine carriers or cold trucks during transportation
A related issue is that when new vaccines are introduced into the existing vaccine regimen
(as is often the case), this often necessitates major additions to existing cold chain systems.
The recurring costs of adding cold capacity and the associated equipment maintenance costs
are a significant concern. Although significant progress has been made with the assistance
of global organizations to strengthen cold chain capacities, in order to provide a sustainable
cold chain, PCDs could prove more favorable than conventional refrigerators.
Different initiatives have urged the development of PCDs for vaccines. One of the best
known was issued in 2009 by PATH (Program for Appropriate Technology in Health)
entitled “Long-term Cold Storage Containers and Carriers for Heat/Freeze-sensitive Vac-
cines: Target Performance Criteria” (McCarney 2009). PATH’s goal was to foster the
creation of PCDs with cold storage times of between one week and one month. The
specific PCD designs sought were either for a 0.5–5 liter capacity mobile device or a
60-liter stationary device. Over the past several years, different designs have emerged
from various sources including Georgia Tech (McClean et al. 2010), the University of
Maryland (Conway et al. 2012) and Intellectual Ventures (Gasperino and Yildirim 2011;
Intellectual Ventures 2013). Examples of commercially-manufactured PCDs described by
PATH include the SAVSU Nano-Q 14 (Savsu 2013), the Dometic RCW 4 (Dometic 2013),
the AirContainer Package System (2013) and the SureEnergy True Chill (TrueEnergy 2013).
Many of these devices have been piloted in different settings for periods of several months
up to one year; e.g., SAVSU’s Nano-Q 14 and SureEnergy’s True Chill (Project Optimize
2013).

1.2 PCD designs and configurations

PCDs are essentially designed to maintain proper temperatures to hold vaccine cool between
replenishments. Many considerations go into the design of PCDs. They must avoid heat
loss under variable ambient temperatures, the design must facilitate easy access to vaccines
stored in them, and they must be constructed to be durable enough to withstand frequent use.
Furthermore, the device must be evaluated through field trials and meet WHO requirements.
Elements of different PCD designs and configurations considered are as follows:
(a) PCD Volume: A larger volume can provide more vaccine storage space. The drawbacks
are increased cost and greater device weight. The volume also depends on the internal
geometry and compartment design of the PCD; these factors influence the vaccine and
ice packing efficiency.
(b) PCD Mass: The total mass of a refilled PCD is based on the weight of the PCD itself, the
vaccines placed inside of it and the ice loaded into it. The total mass impacts the flexibil-
ity associated with handling and delivering. For example, if the total mass is greater than
25 kg, then the device will typically be delivered using a truck (Village Reach 2012).
Since hand-carried movement is the most common way of vaccine delivery at the heath
post level, device weight is an important consideration. Using lighter insulation material
for PCD design can reduce total mass.
Ann Oper Res (2015) 230:87–104 91

(c) Ice Requirement: The heat leakage rate impacts the volume of ice needed to keep the
required cool temperatures. Ice requirement depends on the device geometry, access
point design, device materials and cooling medium.
(d) Ice Buffer: Ice buffers are used to mitigate multiple risks including variation in ambient
temperature, variation in cooling media performance, frequency of device access, vari-
ation in PCD performance, and potential delays in exchanging PCDs and recharging.
Ice buffers require additional space in a PCD and increase the demand for recharging
freezers needed at the recharging points.
(e) Hold Time (HT): Hold time measures the longest time interval that vaccines can be
stored in a PCD. Heat leakage rate and cooling media provided are the primary factors
that determine the HT. In general, hold time is measured using an ambient temperature
assumption (one common choice is +43 °C) to cover most use case situations.
This paper represents the first effort at systematically addressing many of the key issues
affecting PCD design including size, cost, and replenishment interval in order to better guide
the development of PCDs. It examines actual use cases with real data from a representative
country, and presents a model that captures the various tradeoffs that must be considered
in developing a suitable PCD while providing guidance on how to arrive at a good, robust
design.
The remainder of the paper is organized into four sections. Section 2 discusses different
PCD use cases. Section 3 presents a model for evaluating the cost effectiveness of different
PCD designs. Section 4 shows results from this model using data from the country of Niger.
Section 5 provides a summary of the findings and conclusions.

2 Use cases

We next explore the deployment of PCDs using the West African nation of Niger as a pilot
for studying PCD applications. Niger is a typical example of a country where PCDs could
prove useful, with a high birth rate, relatively low vaccine coverage, poor healthcare facil-
ities, and prevalence of infectious diseases. However, our analysis is generalizable to other
countries and to other use cases.
This study investigates using PCDs for delivering and storing vaccines at health posts,
which are typically located at the lowest level of the vaccine distribution chain, and where
vaccines are typically administered. Usually these vaccine supply chains run from a sin-
gle central store, through regions and districts before reaching health posts. Each district
services a number of health posts and we assume that PCDs are recharged at the district
level, i.e., PCDs are filled with vaccines and cooling media (say, ice) and delivered from a
district to the health posts that it services. The charged PCDs contain enough ice to keep
vaccines cool until the next recharging time point. A health post might need more than
one PCD depending upon its volume and the vaccine demand at the health post. The de-
ployed PCDs remain at the health post and are allowed to be re-opened as required dur-
ing vaccination days. Extra PCDs are preserved at the recharging points ready to be filled
with vaccines and cooling materials when they enter the deployment rotation. Recharged
PCDs are brought from the recharging points to the health post and swapped out with
used ones. Unused vaccines left in the used PCDs are returned to the recharging points
with the used PCDs. These vaccines can be restocked and sent to the health posts again
(Fig. 1).
92 Ann Oper Res (2015) 230:87–104

Fig. 1 Passive cold device use

case

Table 1 PCD use case options

Factor Options Reference

Target population Population served at clinics Niger census data

Vaccines
Replenishment frequency
Extra devices for rotation
Transportation mode
EPI
EPI + Rotavirus + Pneumococcal
Every 2, 4, 6, 8 weeks
20 %
Motorcycles or 4 × 4 trucks
(WHO-8 2011; WHO-9 2013)
(WHO-7 2009)
Niger cMYP (WHO-6 2011)
Assumptions
Niger cMYP (WHO-6 2011)

2.1 Use case options

Many competing factors influence PCD designs and configurations. To evaluate whether a
particular PCD design is robust, this study uses Niger as the basis and examines performance
corresponding to several critical factors, some of which might change across different use
cases. Key factors considered are listed below and summarized in Table 1.
(a) Target Population: The target population includes pregnant women and children under
two years of age. The number of children of each age is typically computed by discount-
ing the number of annual newborns by a mortality rate.
(b) Vaccine Schedule: The baseline scenario uses the Expanded Program on Immunization
(EPI) from the WHO. We also consider the introduction of new vaccines.
(c) Replenishment Frequency: The replenishment frequency is typically set in each country
and its value can dictate whether a particular PCD design is feasible, based on how
long the PCD can keep vaccines cold and whether it can store enough vaccine to fulfill
requirements until the next replenishment. In the baseline, each health post is assumed
to receive re-filled PCDs from the recharging point every four weeks, as is usually the
case in practice. We also investigate two-week, six-week and eight-week replenishment
periods.
(d) Transportation Mode: Two transportation modes are considered: motorcycles and 4 × 4
trucks. For both modes we assume that shipments are point-to-point meaning they go
directly from the recharging point to the vaccination location. Motorcycles have two
primary limitations relative to 4 × 4 trucks. First, motorcycles can only carry PCDs with
limited mass, larger PCDs can only be transported by trucks. Second, a motorcycle can
carry at most 2 PCDs (and sometimes only one depending on the size) but a truck can
Ann Oper Res (2015) 230:87–104 93

carry several more (the exact number depends on the size of the PCD but generally it is
8 to 15).
(e) Device Rotation: Extra PCDs in the system are allocated to the recharging points. Hav-
ing additional PCDs incurs additional costs, but the number of additional PCDs affects
the types of replenishment policies that are feasible. For example, if the device rotation
ratio is 100 % (i.e., the system has double the number of PCDs needed across all health
posts), then all used PCDs at the health posts can be exchanged with the extra PCDs at
the same point in time. A strategy of grouping replenishments can reduce traveling dis-
tance by visiting all health points in a single route, and therefore, save on transportation
costs. However, if the health posts are going to be served using multiple smaller routes
then a rotation ratio of less than 100 % can be employed.

3 Model description

We develop computational models to conduct a PCD cost analysis to assess the benefits of
using PCDs in the vaccine supply chain. Table 2 presents our model notation. The mod-
els can be applied to answer questions such as: (1) What is the optimal PCD design for a
particular deployment area? (2) How does changing the vaccine schedule impact the cost
effectiveness of different PCD designs? (3) What is the optimal replenishment frequency?

3.1 Estimating open vial waste

“Open vial waste” refers to the fact that for some vaccines, when a multi-dose vial is opened
during a vaccination session, if the vial is not fully used during the vaccination session
then all of the vaccine remaining in the opened vial must be discarded and cannot be used
on a subsequent vaccination day (WHO-4 2000). A typical example is when a lyophilized
multi-dose vaccine vial is opened and reconstituted with diluent the vaccine is only good
for about six hours; after this time period any left-over vaccine doses must be discarded
(WHO-5 2002). The amount of open vial waste depends on the vial size and the demand
during a vaccination session (Lee et al. 2010). Thus, rural areas that have a small number of
children arriving per vaccination session might need to open more vials per dose of vaccine
administered than urban areas where the population served is larger. This results in higher
open vial waste in rural areas, especially for vaccines that come in larger vials. Open vial
waste (OVW) is a major consideration and estimating OVW is important to ensure that the
correct amounts of vaccines are ordered and vaccine stock-outs are minimized.
Consider the demand di,j of vaccine i at health post j during a vaccination session. For
multi-dose vials i is greater than or equal to two. The OVW oi,j is computed as:
di,j
oi,j = 1 − (1)
di,j /i i
In (1), the denominator of the quantity subtracted from 1 is equal to the number of doses
of vaccine made available (vials opened di,j /i  multiplied by vial size i ). Thus, this
quantity represents the fraction of doses that are actually used, so that oi,j is the fraction
of doses that are not administered to the recipients. Estimates of OVW rates are obtained
by simulating vaccine demands drawn from a Poisson distribution with mean equal to the
average daily demand (see Lee et al. 2010, for additional details).
94 Ann Oper Res (2015) 230:87–104

Table 2 PCD model notation

Sets of indices

Index Description
[I ] Index set of all vaccines, i ∈ [I ]
[J ] Index set of all health posts, j ∈ [J ]

Notation Description

af Annual ice recharging cost

an Annual device cost
as Annual shipping cost
at Annual total cost
cj Vaccine storage space (in liters) per PCD at health post j
di,j Demand (in doses) for vaccine i at health post j
Du Set of all IHCs in district u
ek PCD Ice buffer (%)
en Extra percentage of PCDs for rotation
ev Vaccine vial buffer
f Capacity (in liters) of an ice recharging freezer
g PCD Gross volume (in liters)
hj Required hold time (in days) at health post j
k Ice (in kg) needed to maintain PCD temperature for one week
i Doses per vial for vaccine i
vi Vial volume for vaccine i
mj Total mass (in kg) of a loaded PCD at health post j
nj Number of PCDs required at health post j
oi,j Fraction of open vial loss for vaccine i at health post j
pb Annual energy cost (in $) per freezer
pf Unit cost of ice recharging freezer
pn PCD Unit cost
ps,m Average transportation cost for a district to health post round trip (in $) via motorcycle
ps,t Average transportation cost for a district to health post round trip (in $) via truck
qi,j Number of vials of vaccine i ordered at health post j
rf Ice packing ratio for freezer
rk Ice packing ratio of PCD
rv Vaccine packing ratio of a PCD
v Maximum volume for a truck load (in liters)
w Empty weight (in kg) of PCD (unit: kg)
z Maximum weight for a motorcycle load (in kg)
θj Number of vaccination days in a replenishment interval at health post j
δ Vaccine density (in kg/cm3 )

3.2 Vaccine order quantity

To ensure that vaccines are received in correct quantities, the vaccine ordering decision
involves the assessment of demand forecasts, the target population, desired coverage and
Ann Oper Res (2015) 230:87–104 95

the OVW factor. In addition, vaccine buffers are considered to avoid vaccine stockouts due
to random events such as shipping delays, vaccine spoilage, etc. The ordering buffer is based
on the WHO guideline (Rajgopal et al. 2011).
The order quantity (or more precisely, the order-up-to level) qi,j for vaccine i at location
j that is replenished every θj vaccination days is computed via
 θj di,j  
1−oi,j
qi,j = × (1 + ev ) (2)
i
θ d d
j i,j
The fraction 1−o i,j
(= θj  i,ji i ) is the total number of doses that must be ordered to
cover demand and OVW during a replenishment interval. This quantity is divided by vial
size i and rounded up to obtain the number of vials needed. Finally, the number of vials
ordered is equal to the vials needed inflated by the extra vial buffer, ev , and rounded up
again.

3.3 PCD vaccine storage space

For each PCD, the vaccine storage space is the portion of the internal space available after
the ice has been packed. The volume required for ice is based on the required hold time at
the health post where the PCD is used and how efficient it is in preventing heat leakage; the
latter characteristic depends on the design of the device. A linear relationship is assumed
between the required hold time and the corresponding ice requirement (with k kg required
for a one-week hold time). An additional ice buffer, ek , is used to account for any stochastic
variations in the hold time. The density of ice is assumed to be 0.917 kg/m3 , so that after
accounting for the ice packing ratio, rk < 1, the volume in liters that is occupied by ice is:
khj (1 + ek )
(3)
0.917rk
The remaining space in a device with gross volume g is further discounted by the vac-
cine packing ratio (the fraction of storage volume in a device that can actually be occupied
by vaccine vials), and the effective vaccine storage space cj for a device at location j is
computed as:
 
khj (1 + ek )
cj = rv g − (4)
0.917rk

3.4 Total weight of a fully-loaded PCD

Knowing the vaccine storage space cj , the total weight, mj , of a completely filled device
used at location j is the sum of the PCD weight, the ice weight and the vaccine weight.
Given an average vaccine density of δ, this may be computed as
khj (1 + ek )
mj = w + + cj δ (5)
rk

3.5 The number of PCDs needed for a country

A single PCD type is assumed to be used for vaccine delivery in a country. This is the
preferred approach because it has several benefits, including reduced initial and mainte-
nance costs, and from an operational standpoint, more flexibility with respect to sharing or
96 Ann Oper Res (2015) 230:87–104

exchanging PCDs or parts between different locations. The number of PCDs required for
health post j is:
 
i vi qi,j
nj = (1 + en ) (6)
cj
In (6), the vaccine order quantities qi,j for computing the total volume requirements at
location j (the numerator) come from (2) and the available vaccine storage space per device
cj comes from (4). The parameter en accounts for the extra buffer of PCDs needed during
replenishment; the additional PCDs are used to ensure that each heath post receives re-filled
PCDs when empty PCDs are returned to the recharging locations. The total number of PCDs
needed for a country can then be computed via:
N (ek , en , ev , rk , rv , k, g, Θ, H, D, L, O, V )
θj di,j

 i vi   i,j  × (1 + ev ) 
1−o

= nj = i
khj (1+ek )
(1 + en ) (7)
j j rv [g − 0.917rk
]
In (7), the parameters ek , en , ev , rk , rv , k and g are applied to each health post; other
parameters may be different depending on the health post populations, required hold time
or vaccine characteristics. Θ is the vector of the number of vaccination days over a replen-
ishment interval at the health posts, H includes all hold time settings at health posts, D is a
demand matrix of vaccines i at health posts j , L is a vector of vaccine vial sizes, O is the
matrix of the estimated open vial waste of vaccine i at health post j and V is a vector of
vaccine vial volumes.

3.6 System cost

The system costs include PCD costs, shipping/transportation costs and ice recharging costs.
The PCD unit cost pn is amortized over its useful life, which is assumed to be ten years of
use. With an assumed interest rate of 3 % and a corresponding annual capital recovery factor
of 0.1172, the PCD annual cost an is:
an = 0.1172pn N (ek , en , ev , rk , rv , k, g, Θ, H, D, L, O, V ) (8)
Annual transportation cost is calculated based on the total expected number of round
trips traveled in a year between recharging points and health posts. This is multiplied by an
estimated cost per round trip, ps,m or ps,t , which depends on what sort of vehicle is used
for PCD delivery. The cost per round trip is estimated based on average distance traveled,
vehicle investment, fuel, maintenance and personnel. In addition, we consider vehicle load
limits. A typical motorbike used for EPI vaccine delivery has a capacity of about 15 kg in
the front and 25 kg in the back (Village Reach 2012). For example, if the total mass of a
replenished PCD is 14 kg, then two PCDs can be carried at the same time (one on the front
and one on the back). If the total weight of a PCD is greater than 25 kg, then the shipment
can only be done using 4 × 4 trucks. For the 4 × 4 trucks there is not a weight limit but
rather we assumed a 720 L maximum total PCD gross capacity based on the storage space
available within representative 4 × 4 trucks. With a required hold time of hj the number of
annual shipments is 365hj
for location j . Assuming a maximum load/capacity of z kg or v
liters for motorcycles and trucks, respectively, the number of trips per shipment may then be
computed. Therefore the annual expected transportation cost is calculated as:
  nm
ps,m j  j z j  365 for transport by motorcycles
as =  nj g 365hj (9)
ps,t j  v  hj for transport by trucks
Ann Oper Res (2015) 230:87–104 97

The annual cost of recharging ice is calculated based on the recharging freezers needed,
along with their maintenance and energy costs. The number of recharging freezers needed
is calculated based on the ice volume needed at the health posts. Assuming a freezer with
ice making capacity of f liters, a freezer packing factor of rf , a useful freezer life of ten
years, a 3 % interest rate, and a maintenance cost of 5 % of the purchase price in each year,
the total annual cost may be computed as:

 j ∈Du 0.917rf 
nj khj (1+ek )

af = (0.1172pf + 0.05pf + pb ) (10)

u
f
where Du is the set of all IHCs in district u and u is summed overall all districts.
In (10), the first term within the parentheses is the annualized cost of the freezer purchase
price, the second term is the annual maintenance cost, and the third term is the annual energy
cost per freezer; the term outside the parentheses is the number of freezers needed. Finally,
the total system cost is:
at = an + as + af (11)

4 Results

In order to compare the economic impacts of different possible PCD designs, the analysis
in this section is based on actual data from Niger that was obtained from WHO documents
(WHO-6 2011), GAVI data (GAVI-2 2012) and from health officials in Niger. There are
42 district stores which recharge PCDs to service the health posts for which they are re-
sponsible. The average number of health posts within a district is 15, ranging from 5 to
35. The values of the PCD cost parameters for various weights, volumes and ice require-
ments are shown in Table 4. The weight and volume values are similar to storage devices
presently available from Savsu (http://www.savsu.com/savsuproducts.html) and Dometic
(www.dometic.com/enlu/Europe/Luxembourg/Medical-Systems/Cold-Chain).
The cost values are based on information about similar types of devices and reflect that
fact that larger devices tend to be more expensive, and that a design requiring less ice per
week of hold time tends to be more expensive (to reflect higher material costs, superior in-
sulation, or higher fabrication costs for obtaining special geometries). Thus, as an example,
Tables 3 and 4 indicate that a 30 liter device that weighs 12 kg and requires 2 kg of ice for
a week of hold time would cost 350 + 750 + 750 = $ 1,850. We use the cost/FIC to assess
the performance of different PCD designs over a time horizon of one year. The system costs
are calculated using Eq. (11).

4.1 Base case

The vaccine target coverage is based on the GAVI plan for 2015. We selected 2015 because
Niger is scheduled to introduce additional vaccines into the EPI regimen by 2015 and it
would be of interest to see how sensitive a particular PCD design might be to changes in
the vaccine regimen. To calculate the birth cohort for the base case we took the 2011 birth
cohort (GAVI-1 2011) and applied an annual population growth rate of 3.63 % (CIA 2012) to
obtain the projected birth cohort for 2015. We then made that assumption 85 % of this birth
cohort would need to be vaccinated at health posts (with the remainder being vaccinated
through outreach programs or not being vaccinated at all). This results in a birth cohort of
approximately 700,000 children for the base case. The age coverage includes children under
98 Ann Oper Res (2015) 230:87–104

Table 3 PCD costs

contributions for different weight Weight Volume Cost parameter
and volume combinations (kg) (Liters) Weight Volume

7 10 $ 200 $ 250
7 20 $ 200 $ 750
7 30 $ 200 $ 1,250
12 20 $ 350 $ 250
12 30 $ 350 $ 750
12 40 $ 350 $ 1,250
17 30 $ 500 $ 250
17 45 $ 500 $ 750
17 60 $ 500 $ 1,250

Table 4 PCD costs contributions

for different amounts of ice
needed per week of hold time
Ice needed (kg/wk) Cost parameter
1 $ 1,250
2 $ 750
3 $ 250

two years of age in accordance with the Niger EPI program. The hold time is set as four
weeks, and this assumption is based on the current ordering policy in Niger. We use three
numbers, volume-weight-ice, to present different PCD designs. The first number refers to
the gross PCD volume (g), the second to the unloaded PCD mass (w), and the third to the
amount of ice required to attain a hold time of 1 week (k).
Table 5 shows the results of cost/FIC for different PCD designs. This metric is obtained
by dividing the total system cost at by the number of children who are fully immunized each
year. The blank values indicate that the designs are infeasible because the ice requirements
for the four-week hold time are greater than the gross volume of the PCD (thus leaving no
room for vaccine storage). Designs that require less ice are generally more cost effective
because this leaves more room for vaccine storage and results in fewer PCDs being needed,
and also because the ice recharging cost is lower. There is an exception when the PCD has
weight 17 kg and a volume of 60 L. For this weight and volume configuration the 17-60-2
device has lower cost ($ 1.09) than the 17-60-1 ($ 1.11) because with this large volume,
most health posts require only one PCD and the 17-60-2 device price is cheaper than then
the 17-60-1. However, it is interesting that the 17-60-3 device becomes more costly even
though it has the lowest device price. This is because while the device price plays a major
role, the higher cost of recharging ice negatively impacts the overall cost performance.
The optimal PCD design corresponds to the 17-45-1 design. This is because for this
design most health posts require a single PCD and the PCD storage space is highly utilized.
In addition, the ice recharging cost is the least since each PCD only requires 1 kg of ice per
week of hold time.

4.2 Impact of introducing new vaccines

Table 6 illustrates similar information but with the vaccine demands based on adding new
vaccines to the current EPI regimen. Two vaccines that are currently recommended for in-
clusion by 2015 are the Rotavirus and Pneumococcal vaccines. A major issue is that these
Ann Oper Res (2015) 230:87–104 99

Table 5 Cost/FIC of different

PCD designs in base case PCD weight PCD volume Ice needed (k kg/week of hold time)
3 kg 2 kg 1 kg

7 10
20 $ 1.33
30 $ 1.72 $ 1.08
12 20 $ 1.64
30 $ 1.71 $ 1.08
40 $ 1.59 $ 1.23 $ 1.06
17 30 $ 1.51 $ 1.19
45 $ 1.36 $ 1.10 $ 1.05
60 $ 1.12 $ 1.09 $ 1.11
Base case: hold time = 4 weeks

Table 6 Cost/FIC of different

PCD designs in the new vaccine PCD weight PCD volume Ice needed (k kg/week of hold time)
introduction case 3 kg 2 kg 1 kg

7 10
20 $ 3.35
30 $ 4.51 $ 2.52
12 20 $ 4.29
30 $ 4.49 $ 2.52
40 $ 3.23 $ 2.10 $ 1.75
17 30 $ 3.01 $ 1.99
45 $ 2.56 $ 1.91 $ 1.61
60 $ 1.90 $ 1.66 $ 1.62

vaccines will take up significant space: three doses of each of these will be required in the
EPI regimen and the volume per dose of the Rotavirus vaccine is especially large compared
to the others. As Table 6 illustrates, the cost/FIC rises dramatically in the cases where the
devices have small vaccine storage space (such as 7-20-1, 7-30-2, 17-20-1, 17-30-2 and 17-
40-3). To satisfy the greater demands on vaccine volumes, larger PCDs are more desirable.
Thus, larger PCD volumes have less of a cost increase for the new vaccine introductions.
For example, the 17-60-1 design costs $ 1.11/FIC in the base case and then increases up
to $ 1.62 for the new vaccine introductions. This gap is much smaller than for small PCD
designs (such as 7-20-1, which increases in cost from $ 1.33 to $ 3.35).
The variations between different designs also become greater with new vaccine introduc-
tions; the lowest cost (17-45-1) is about three times cheaper than the highest cost design
(7-30-2). It is also interesting that the 17-45-1 is the most inexpensive design in both the
baseline and the new vaccine introduction cases.
Comparing the system costs of the three ice requirement options, the 1 kg designs are
consistently lower than the 2 kg and 3 kg designs across different weight and volume set-
tings. Unlike in the base case, the 17-60-1 design is less expensive than the 17-60-2 design,
as the designs requiring less ice have more vaccine storage space, which is important when
introducing new vaccines and therefore reduces the number of PCDs needed in the system.
100 Ann Oper Res (2015) 230:87–104

Table 7 Sensitivity analysis: Different HTs for the 17-45-1 PCD

Parameter Cost/FIC Total cost

PCD cost Transportation cost Ice recharging cost

2 week HT $ 0.39 $ 1.16 $ 0.04 $ 1.60

4 week HT $ 0.41 $ 0.47 $ 0.06 $ 1.05
6 week HT $ 0.62 $ 0.39 $ 0.12 $ 1.13
8 week HT $ 0.83 $ 0.29 $ 0.19 $ 1.31

4.3 Sensitivity analysis of different hold times

The purpose of this analysis is to verify if there is a robust device design for different hold
time settings. The HT duration selected is related to the replenishment frequency. Ice re-
quirements are calculated to ensure that the PCD will maintain proper cooling for the dura-
tion of the time between replenishments.
Table 7 shows the cost summary for the 17-45-1 PCD, which was the most cost-effective
design in the two previous subsections. The PCD cost increases with the HT duration, from
$ 0.39 (2 weeks) to $ 0.83 (8 weeks); these costs are more for longer HTs since each PCD
requires more ice and has less storage space and therefore more PCDs may be required.
Transportation cost is the highest for the two week HT due to more frequent shipping. There
is a trade-off between transportation and ice recharging costs, the two week HT has the
lowest ice recharging cost as it requires less ice volume. The system cost is the lowest when
the HT is 4 weeks, and the system cost increases for either shorter or longer HTs.

4.4 Sensitivity analysis of PCD costs parameters

Suppose the cost contributions from the three design parameters of weight, volume and
ice requirements (and hence the device price) are not known with certainty. The impact of
variations in the device price with different contributions from the three parameters can be
explored using sensitivity analysis. In Table 8 “regular” refers to the original value of the
cost contribution for the level of the design parameter in question (i.e., the corresponding
value in Tables 3 or 4), while “low” and “high” refer to where this value is reduced or
increased respectively by 50 %. As an example, for the 12-30-1 device, “low,” “regular,”
and “high” would use, respectively, values of 175, 350 and 525 for the contribution to total
cost from the weight parameter, 375, 750 and 1125 from the volume parameter, and 625,
1250 and 1875 from the ice requirement parameter.
Table 8 summarizes the most economical designs for different cost parameter combina-
tions. The 17-45-1 design is always the best when the ice requirement parameter is at the
regular level and almost always the best when this parameter is at the high level. However,
if this parameter is at its low value then the medium volume and medium weight device
(12-30-1) becomes the best option. It is also important to note that combining changes in
multiple factors impacts which design is best. One example is if the cost contribution from
the ice requirement parameter is at the high level and the contribution from the volume
parameter is at the low level, the 17-60-3 design is the lowest cost design.
Because actual costs of different PCD features might differ from the ones assumed in
Tables 3 and 4, we perform a sensitivity analysis to determine if there is a design that is
robust with respect to these differences. We select each of the three PCD designs in Table 8
(12-30-1, 17-45-1, and 17-60-30) that turn out to be optimal for some specific combination
Ann Oper Res (2015) 230:87–104 101

Table 8 Optimal PCD designs

Weight parameter Volume parameter Ice requirement parameter

Low (−50 %) Regular High (+50 %)

Low (−50 %) Low (−50 %) 12-30-1 17-45-1 17-60-3

Regular 12-30-1 17-45-1 17-45-1
High (+50 %) 12-30-1 17-45-1 17-45-1
Regular Low (−50 %) 12-30-1 17-45-1 17-60-3
Regular 12-30-1 17-45-1 17-45-1
High (+50 %) 12-30-1 17-45-1 17-45-1
High (+50 %) Low (−50 %) 12-30-1 17-45-1 17-60-3
Regular 12-30-1 17-45-1 17-45-1
High (+50 %) 12-30-1 17-45-1 17-45-1

of cost parameters, and then evaluate it across each of the 27 scenarios in the table to see
far it deviates from the optimal solution to the scenario. Different robustness criteria can
be used to determine the overall most robust solution: two of the most common criteria
are the average regret and min-max regret. Average regret for each PCD design is obtained
by first computing the deviation of its cost from the optimal cost for each scenario, and
then averaging these deviations across all scenarios. The best design with respect to this
criterion would be the one with the lowest average deviation. On the other hand, min-max
regret evaluates the maximum deviation from the optimum cost across all scenarios for each
design, and prefers the one that minimizes the maximum deviation (Kouvelis and Yu 1997).
The 12-30-1 design shows zero deviation in the low-cost (−50 %) ice requirement cases,
2 % to 3 % cost deviations for the regular-cost ice requirement cases and the worst cases
occur at the high-cost ice requirement, which has cost deviations ranging from 7 % to 14 %.
The 17-45-1 design shows 1 % to 5 % cost deviations in the low-cost (−50 %) ice re-
quirement cases, zero deviation for the regular-cost ice requirement cases and 0 % to 5 %
deviations in the high-cost ice requirement cases. The 17-60-3 design shows 13 % to 19 %
cost deviations in the low-cost (−50 %) ice requirement cases, 1 % to 10 % deviation for the
regular-cost ice requirement cases and zero deviation in the high-cost ice requirement cases.
The 17-45-1 PCD is the most robust solution with regard to cost parameter variations using
both the average and min-max regret criteria because its average regret is 1 % compared to
4 % for the 12-30-1 design and 8 % for the 17-60-3 design and its maximum regret is 5 %
compared to 14 % for the 12-30-1 design and 19 % for the 17-60-3 design.

4.5 Generalization of PCD designs

It has been noted in the previous discussion that many parameters affect PCD design and
cost performance. Uncertainty about the population, and hence for vaccine demand, is one of
the main factors affecting PCD design. This led to an additional analysis based on changing
the vaccine demand to determine how it affects the choice of PCD design. The annual birth
cohort being served at a health post was set at levels of 500, 1000, 1500, 2000, 2500, and
3000. For each level, we assumed there were 16 health posts serving an annual birth cohort
of that size within one district location (on average, there are approximately 16 health posts
per district in Niger). Two vaccine schedules were tested—the current Niger EPI schedule
and with the introduction of Rotavirus and Pneumococcal vaccines. Table 9 lists the optimal
PCD designs for the six different catchment populations for both vaccine schedules. For the
102 Ann Oper Res (2015) 230:87–104

Table 9 Optimal PCD designs

for different catchment sizes Annual birth per IHC Vaccine
EPI New vaccine introduction

500 12-30-1 12-40-1

1,000 17-45-2 17-45-1
1,500 12-40-1 17-60-2
2,000 17-45-1 17-45-1
2,500 17-60-2 17-60-1
3,000 17-60-1 17-60-1

smallest catchment population (i.e., 500 annual births at each IHC), 12-30-1 is the most cost
effective design in the EPI only case, and 12-40-1 is the best design for the new vaccine
introductions. For the medium catchment population (i.e., annual births are either 1000 or
2000), the best designs are 17-45-1 and 17-45-2 (where 17-45-1 is also the best design in
Niger). For the larger catchment population the larger designs perform better and the 17-60-
1 design is the optimal design in both the base EPI and new vaccine introduction cases when
there are 3,000 annual births.
In general, there are several advantages of having a single, standard PCD type in a
country—these were enumerated earlier on at the beginning of Sect. 3.5. However, Table 9
clearly indicates that the volume of the optimal design increases as the catchment area size
increases. Thus, it might be best to actually use multiple sizes in a country that has health
posts with very different catchment size areas, despite the reduction in flexibility and op-
erational simplicity. Performing a robustness analysis based on the size of the catchment
population (similar to that done for Niger) indicates that the 17-45-1 design (which was the
most robust device in Niger) and the 17-60-1 design are the most robust across the tested
range of catchment sizes. Such analyses could easily be extended to other ranges of catch-
ment populations and help to identify the best designs for those use cases.

5 Summary and conclusions

In this study, we develop a computational model to evaluate different PCD designs for vac-
cine distribution for various use cases. We provide an example to demonstrate how PCDs
can be used to deliver and store vaccines, and evaluate the proposed use-cases employing
actual data from the system in Niger. The cost analysis provides information to understand
the benefits of using PCDs and help inform product development engineers at PCD man-
ufacturing companies about what designs would be the most cost-effective. Our sensitivity
analysis identifies a robust design for the situation when device price cannot be known with
accuracy. Similar analyses can be conducted for deviations of either transport cost or ice
recharging cost. Our general model provides the required resources and their associated
costs for vaccine transportation and storage. The model can be readily populated and re-
used for other countries, which is useful for vaccine logistics and economic modelers to
assess the cost-effectiveness of PCDs.
There are limitations to this study. We examined the cost tradeoffs in changing either
PCD volume or the ice needed per week of HT parameters, in the absence of changing the
weight parameter. This is because in the analysis presented here there is no cost penalty as-
sociated with device weight, and therefore heavy, but low price, designs are always the best
choice. However, in real life device weight must be considered along with its impacts on
Ann Oper Res (2015) 230:87–104 103

logistics system design, particularly aspects of loading/unloading operations, vehicle max-

imum loads, and how device portability affects outreach vaccinations. As an example, rea-
sonable design considerations might stipulate that at most 25 kg may be moved by a person,
a 50 kg device would require two people (and hence, higher logistics costs), and a device of
more than 50 kg would be primarily stationary and could not be used for outreach activities.

Acknowledgements This study was supported by the Vaccine Modeling Initiative (VMI), funded by the
Bill and Melinda Gates Foundation, and the National Institute of General Medical Sciences Models of Infec-
tious Agent Study (MIDAS) grant 1U54GM088491-0109. The funders had no role in the design and conduct
of the study; collection, management, analysis and interpretation of the data; and preparation, review, or
approval of this manuscript.

References

AirContainer Package System (2013). Pre-qualified PCM package system for temperature-sensitive.
http://www.aircontainer.com/the-product.aspx. Accessed July 10, 2013.
CIA (2012). The world factbook: Niger. https://www.cia.gov/library/publications/the-world-factbook/geos/
ng.html. Accessed July 10, 2013.
Conway, M., et al. (2012). Improving cold chain technologies through the use of phase-change material. Uni-
versity of Maryland, College Park. http://drum.lib.umd.edu/bitstream/1903/12491/1/FRESH.pdf. Ac-
cessed July 10, 2013.
Dometic (2013). RCW 4. http://www.dometic.com/enlu/Europe/Luxembourg/Medical-Systems/Cold-Chain/
Transport-Boxes/products/?productdataid=79144. Accessed July 10, 2013.
Gasperino, D., & Yildirim, O. (2011). Design and prototyping of a passive cold chain vaccine storage device
for long hold times. In Comsol conference: intellectual ventures, Boston, MA.
GAVI-1 (2011). GAVI alliance: country hub. http://www.gavialliance.org/country/. Accessed July 10, 2013.
GAVI-2 (2012). Niger: GAVI alliance. http://www.gavialliance.org/country/niger/. Accessed July 10, 2013.
Intellectual Ventures (2013). Vaccine cold chain device. http://www.intellectualventures.com/index.php/
inventions-patents/our-inventions/vaccine-cold-chain-device. Accessed July 10, 2013
Jamison, D., Breman, J., Measham, A., Alleyne, G., Claeson, M., Evans, D., Jha, P., Mills, A., & Musgrove,
P. (2006). Disease control priorities in developing countries (2nd ed.). New York: Oxford University
Press.
Kaufmann, J. R., Miller, R., & Cheyne, J. (2011). Vaccine supply chains need to be better funded and strength-
ened, or lives will be at risk. Health Affairs, 30(6), 1113–1121.
Kouvelis, P., & Yu, G. (1997). Robust discrete optimization and its applications. New York: Kluwer Aca-
demic.
Lee, B. Y., Norman, B. A., Assi, T. M., Chen, S. I., Bailey, R. R., Rajgopal, J., Brown, S. T., Wiringa, A.
E., & Burke, D. S. (2010). Single versus multi-dose vaccine vials: an economic computational model.
Vaccine, 28(32), 5292–5300.
Laxminarayan, R., & Ganguly, N. K. (2011). India’s vaccine deficit: why more than half of Indian children
are not fully immunized, and what can—and should—be done. Health Affairs, 30(6), 1096–1103.
McCarney, S. (2009). RFP for new insulated containers and carriers for vaccine. http://www.technet-21.org/
index.php/component/content/article/95-articles/forums/1500-topic1533. Accessed July 10, 2013.
McClean, K., Melgen, S., Poole, J., Shroff, P., & Wiggins, C. (2010). Thinking outside the (cool) box. Georgia
Tech, Atlanta. http://www-old.me.gatech.edu/jonathan.colton/detm/coolboxfinalreport.pdf. Accessed
July 10, 2013.
Project Optimize (2013). Unplugged and keeping cool—testing off-grid vaccine storage solutions in
Vietnam. PATH/WHO: Seattle/Washington. http://www.path.org/publications/files/TS_opt_unplugged_
viet_br.pdf. Accessed July 10, 2013.
Rajgopal, J., Connor, D. L., Assi, T. M., Norman, B. A., Chen, S. I., Bailey, R. R., Long, A. R., Wateska,
A. R., Bacon, K. M., Brown, S. T., Burke, D. S., & Lee, B. Y. (2011). The optimal number of routine
vaccines to order at health clinics in low or middle income countries. Vaccine, 29(33), 5512–5518.
Savsu (2013). Nano-Q 14 series. http://www.savsu.com/savsunano-q.html. Accessed July 10, 2013.
Stack, M. L., Ozawa, S., Bishai, D. M., Mirelman, A., Tam, Y., Niessen, L., Walker, D. G., & Levine, O. S.
(2011). Estimated economic benefits during the ‘decade of vaccines’ include treatment savings, gains in
labor productivity. Health Affairs, 30(6), 1021–1028.
TrueEnergy (2013). Sure chill. http://www.trueenergy.com/surechill.aspx. Accessed July 10, 2013.
Village Reach (2012). Personal communication with Village Reach, April 2012.
104 Ann Oper Res (2015) 230:87–104

WHO-1 (2009). State of the world’s vaccines and immunization (3rd ed.). World Health Organization. http://
whqlibdoc.who.int/publications/2009/9789241563864_eng.pdf. Accessed July 10, 2013.
WHO-2 (2006). Temperature sensitivity of vaccines. World Health Organization. http://whqlibdoc.who.int/
hq/2006/WHO_IVB_06.10_eng.pdf. Accessed July 10, 2013.
WHO-3 (2012). PQS catalogue > prequalified Devices and equipment. World Health Organization. http://
apps.who.int/immunization_standards/vaccine_quality/pqs_catalogue/categorylist.aspx?cat_type=device.
Accessed July 10, 2013.
WHO-4 (2000). The use of opened multi-dose vials vaccine in subsequent immunization sessions. World
Health Organization. https://apps.who.int/vaccines-access/vacman/mdvp/WHO00.09doc.pdf. Accessed
July 10, 2013.
WHO-5 (2002). Getting started with vaccine vial monitors. World Health Organization. http://whqlibdoc.
who.int/hq/2002/WHO_V&B_02.35.pdf. Accessed July 10, 2013.
WHO-6 (2011). Comprehensive multi-annual plan: expanded program on vaccination, Niger. World Health
Organization. http://www.who.int/immunization_financing/countries/cmyp/Niger_cMYP_doc.pdf. Ac-
cessed July 10, 2013.
WHO-7 (2009). Introduction of rotavirus vaccines into national immunization programmes: manage-
ment manual, including operational information for health workers. http://whqlibdoc.who.int/hq/
2009/WHO_IVB_09.09_eng.pdf. Accessed July 10, 2013.
WHO-8 (2011). Vaccine management and logistics: vaccine volume calculator. http://www.who.int/
immunization_delivery/systems_policy/logistics/en/index4.html. Accessed July 10, 2013.
WHO-9 (2013). WHO vaccine-preventable diseases: monitoring system: 2013 global summary. http://apps.
who.int/immunization_monitoring/globalsummary/countries?countrycriteria[country]=NER. Accessed
July 10, 2013.