Communications and Network, 2013, 5, 4-9
doi:10.4236/cn.2013.53B1002 Published Online August 2013 (http://www.scirp.org/journal/cn)
Copyright © 2013 SciRes. CN
The Internet of Things and Next-generation Public Health
Robert Steele, Andrew Clarke
Discipline of Health Informatics, the University of Sydney, Sydney, Australia
Email: robert.st firstname.lastname@example.org, andrew.clar email@example.com
Received June, 2013
The Internet of things has particularly novel implications in the area of public health. This is due to (1) The rapid and
widespread adoption of powerful contemporary Smartphone’s; (2) The increasing availability and use of health and
fitness sensors, wearable sensor patches, smart watches, wireless-enabled digital tattoos and ambient sensors; and (3)
The nature of public health to implicitly involve co nnectivity with and the acquisitio n of data in relation to lar ge num-
bers of individuals up to populatio n scale. Of particular re levance in rela tion to the Internet of T hings (IoT ) and public
health is the need for privacy and anonymity of users. It should be noted that IoT capabilities are not inconsistent with
maint ainin g priva cy, d ue to t he focus of public health on aggregate data not individual data and broad public health in-
terventions. In addition, public health information systems utilizing IoT capabilities can be constructed to specifically
ensur e p r ivacy, sec ur it y a nd a no ny mity, a s has been developed and evaluated in this work. In this paper we describe the
particular characteristics of the IoT that can play a role in enabling emerging public health capabilities; we describe a
privacy-preserving IoT -based public health information system architecture; and provide a privacy evaluation.
Keywords: Inter net of Things ; Public Health; Population Health; Privacy; Anonymity
The Internet of things can find particular applicability
in the area of public health. This is because public health
is a field where communication with large numbers of
individuals is implicitly required, either for data capture
or public health intervention. In addition, many of the
data inputs required for public health infor mation capture
are increasingly available via the proliferation of con-
sumer heal th and fitness sensors .
The recent rapid growth in both the capabilities and
uptake of mobile devices or Smartphone’s capable of
acting as sensor platforms has the potential to enable a
new generation of public health information systems.
While increasingly, mobile devices and sensors are used
as a tool for individual health data capture, tracking and
feedback, the use of such technologies has not to-date
substantiall y extended into use for public health purposes.
In addition, the use of sensors for individual fitness and
health tracking does not as critically require an IoT infra-
structure as does public health, as individual fitness
tracking does not necessarily require widespread inter-
connectivity between many sensors and processors –
such i ndi vid ual fitne ss a nd he alth d ata onl y str ict ly nee d s
to be made available to the individual user.
In this paper we describe how an IoT-based architec-
ture can be utilized for population health data capture and
public health intervention whilst still maintaining strong
privacy and anonymity for all participating individuals.
Prior work in relation to achieving privacy and security
has relied on a trusted data collector or aggregation
process, whereas our approach does not assume this. In
addition, interestingly, the case for public health usage
doesn’t require the same level of precise data that would
often be required in other IoT applications. For example
the exact location and time of a measured sensor value is
less important than the aggregate value over a period of
time or the trend of change for a mass of people or
community. Public health interventions , are a key
component of future Health Participatory Sensing
Networks (HPSNs) , and in our approach we describe
capabilitie s whereby a tar geted p ublic health i nterventio n
can be distributed, performed and evaluated without the
need for id entif ying d etails o f a n indivi dual t o eve r leave
their mobile device.
Also central to our approach is an anonymizng layer 
within the IoT-based architecture, which utilizes either a
MIX net wor k [ 3 ] or O nio n rout i ng [ 4] . Thi s anonymizing
layer is one of the mechanisms to enforce priva-
cy-preserving public health-related communications.
In Section 2 we discuss the relevant emerging capabil-
ities of the IoT and their relevance and match to public
R. STEELE, A. CLARKE
Copyright © 2013 SciRes. CN
health goals. In Section 3, we describe the IoT-based
public health information system architecture, including
the resultant data capture and public health intervention
capabilities, in Section 4 we describe privacy and ano-
nymity and in Section 5 provide a privacy evaluation.
Thi s is foll owe d by the Conclusion.
2. The Internet of Things and Public Health
In this section we identify novel IoT capabilities and
overview their relationship to public health measures.
Many public health measures can already be captured
automatically via such Io T capab ilitie s.
2.1. Internet of Things Health Sensor
The proliferation of commercial fitness and health
sensors provides new mechanisms for population health
data capture. Emerging sensors also already able to cap-
tu re ma n y bio medical measures captured in public health
data surveys. In addition, these have a number of charac-
teristics quite distinct from traditional survey—based
population health data capture approaches.
• Larger participant numbers
• More detailed data
• Captured electronically
• Direct measurement, not human response
The area of IoT personal health sensor and software
development  is one of the most active areas of the
IoT ecosystem. This is possibly due to the relevance of
these individual sensors to both consumer-centric phone
technologies and the increasing interest to leverage such
technologies for improved personal wellness, health and
Fitness and Ac tivity Senso r s
Commercial implementations such as Nike Fuel and
Jawbone Up  demonstrate the achievability and poten-
tial for continuous ph ysical activity sensing. Ja wbone Up
extends beyond physical activity monitoring to include
sleep pattern and quality, and a nutritional diary. Other
well-known examples of such sensors include FitBit,
RunKeeper, myFitnessPal, Pebble Watch, the Basis
Wat ch and Goo gle Gl as s. Suc h fit ness a nd he alt h sen sor s
are the most contemporarily available component of the
IoT that can be utilized for public health as such sensors
are already achieving widespread interest and adoption.
Also of significant relevance is Google Now’s, Activ-
ity Summary  which provides a monthly estimate of
how far an individual has walked and cycled, and comes
as part of Google Android – hence is already extremely
Vital Signs Sensors
Smartwaches such as the Mio Active are able to cap-
ture heart rate. The Amiigo wristband captures blood
oxyge n level s, S o ma xi s provides ECG and EMG sensors
and the mc10 stretchable electronic tattoo can transmit
heart rate and brain activity . The capturing of vital
signs is often more beneficial for individual health care,
but it also add s ne w capab ilitie s to pub lic heal th in for ma-
tion systems. As another example, the Sense A/S moni-
toring patch is able to measure blood pressure .
Blood Constituent Sensors
Increasingly there are wireless-enabled patch technol-
ogies emerging that may be able to capture the levels of
some blood constituents. Examples include the Sano In-
telligence  wearable patch which is touted to allow
the capture of blood glucose and potassium levels, with
further blood constituent capture planned to be forth-
coming. Numerous continuous blood glucose monitoring
systems are also currently available.
Such sensor capabilities in a cheap and accurate form
have the p otential to revol utionize ind ividual health car e,
early detection and preventative health; and also public
It should be noted that such capabilities may be so
beneficial in terms of individual health monitoring and
health maintenance that they could achieve wide adop-
tion. If so, their possible role in public health data cap-
ture will also be proportionally significant.
Other initia tives such as Riderlog  and the Copen-
hage n Whee l [1 2] ar e movi ng to wards c aptur ing p hysi ca l
activity levels and at the sa me ti me, additiona l co ntext ual
data. The Copenhagen wheel goes beyond physical activ-
ity sensing, to urban environment monitoring with air
quali ty and no ise se nsors inc luded in the impl ementa tion
to provide additional data beyond just the activity of the
2.2. Public Hea lth Mea sur es
The various types of data that can be collected via the
above-mentioned IoT sensors, already relate to a majori-
ty of p ublic healt h mea s ures:
• Physical Activity Levels – This is one of the most
important lifestyle factors for chronic health conditions
and other health risks . This can now be quite accu-
rately captured with already available sensors and even
in-built Smartphone capabilities .
• Caloric Burn and Caloric Intake – Caloric burn in-
formation can be captured by a range of activity sensors
as described, and caloric intake can also be increasingly
automatically captured .
• Nutritional Data – As mentioned wearable patches
have the ability to measure potassium levels, one of the
markers of nutrition status [1 5].
R. STEELE, A. CLARKE
Copyright © 2013 SciRes. CN
• Blood Pressure – Blood pressure is a public health
marker of cardiovascular disease . As described,
blood pressure can be captured via a wearable patch such
as the Sense A/S.
• Blood Glucose – a marker of diabetes  can be
captured by wearable patches and other continuous glu-
cose monitoring (CGM) devices.
• Body Mass Index (BMI) – Height is roughly inva-
riant for adults and Bluetooth-enabled scales are increa-
• Sleep Pattern and Regularity – Sleep patterns are
both an indicator and a preventative/risk factor for a
number of conditions. Sleep quality can be captured by
currently available wristband sensors.
3. A Proposed Internet of Things-based
Public Health Information System
The preceding section indicates that even current IoT ca-
pabilities have a significant match to many public health
measure s of interes t. We no w describe a privacy-preserving
public health in formation system architecture.
The overall system architecture (Figure 1) involves one
or many central Servers that communicate with mobile
devices through a MIX network or Onion routing net-
work to provide communications anonymity, and mobile
devices that incorporate local processing and privacy
thresholds to maintain data anonymity/privacy/de-iden-
There are two primary data transmissions from and to
the Server respectively: 1) data requests and public
health interve ntion s are d istributed fro m the server ; and 2)
anonymized data collection submissions are sent to the
server. The core functionality components of the public
health system’s Server are Data Aggregation, Analysis
and Intervention/Data Requests. The Server interfaces
with Public Health Groups, which could include state or
federal health departments, p ublic health researc h institu-
tions or other p ub lic health organizations.
The fundamental architecture can support different le-
vels of data collection and/or potentially public health
intervention, depending largely on the capabilities of the
end user’s mobile devices and preferences of the indi-
vidual users of these devices. As described in the pre-
vious section these IoT capabilities range from those
built into Smartphone to wristband sensors, smartwatc hes,
wearable patches and tattoos and other sensors.
3.2. Anonymous Pub lic Health Data Capture
We have developed our approach such that it does not
require a fully trusted server-based approach that would
Figure 1. Interne t of Thi ngs-based Publi c He a lth Informatio n
likely prove impractical on population-scale applications.
Instead it utilizes an architecture incorporating an anony-
mou s communications network (MIX network or Onion
routing) in combination with de-identification of data
submitted, to provide anonymous submission/inter- ac-
tion. However, this alone would still allow the risk of
re-identification b ased o n qua si-ide nt i fie rs, i n t he form o f
information known about individuals outside the system
that could be u sed to match w ith and re -ide ntify the sub-
mitted data. The most common approach to address this
type of risk is to use a trusted server or aggregation point
to combine and obfuscate data to the point where
k-anonymity  is provided, such that any individual is
indiscernible from k other records based on qua-
To provide an approach that doesn’t require a trusted
server component we propose that a suitable level of
anonymity can be provided by locally processing on the
user’s mobile device collected data into an aggregated
generalized form that can still meet the purposes of pub-
lic health data collection, as described in our previous
work . By utilizing quasi-identifier scores (QIS) and
a threshold approach to privacy limits, the level of pri-
vacy disclosure an individual agrees to can be easily
managed without requiring a case-by-case approval. Ad-
ditionally, our approach involves the specification of and
weighting of the data to be submitted to allow the local
device to alter the resolution and breadth of data submit-
ted to preserve privacy and anonymity, while still sub-
mitting the data needed for public health data purposes.
R. STEELE, A. CLARKE
Copyright © 2013 SciRes. CN
3.3. Anonym ou s Pub lic Health Intervention
A major area of potential usefulness of such an IoT-
based public health information system is the ability to
distribute targeted or personalized public health interven-
tions to individuals.
Additionally, it seems likely that there will be a num-
ber of public health groups (Figure 1) that would be in-
terested in participating in these types of networks and
with individuals able to subscribe or opt-in to partake in
passive or active participation with each such group.
We propose a novel approach in relation to public
healt h i nt er ve nt ion. In l i ne wit h t he lo c al aggr e ga tio n and
processing approach to preserve privacy when submitting
sensing data, it appears appropriate to use a similar ap-
proach for communication from the public health body to
the individual. This novel approach broadcasts larger
generalized public health intervention packages from the
Server to the entirety or subsets of the participants and
then based on local processing, the correct information is
displayed or auctioned on individual devices.
This would allow for communication with individuals
that could be meaningful and personalized without risk
of r e-identi fic at ion o f the ind i vid ua l. This ap p ro a ch c o uld
also be used for the dissemination of micro-surveys to
individuals for additio nal human-entered data collection.
However, this increases the overhead of data distri-
buted to individuals since in all cases the data required
for local processing would need to be received rather
than specific targeted data for each individual.
To improve the flexibility of this approach, we pro-
pose a technique using verification objects that incorpo-
rates a granular approach to hashing and digital signing
of distributed content, by including timestamps and ex-
piry rates to ensure the quality of the distributed data
without direct communication to the associated public
health groups. Our previous work found through imple-
mentation that user CPU and data overheads for this type
of implementation can be quite minimal , without
significant additional overheads for the data owners /dis-
This approach would additionally allow for dissemina-
tion and retrieval of data through the anonymous com-
munication network with users retrieving policy updates
and interventions relevant to them without breaching
4. Privacy and Security
Our system, by applying granular restrictions on data
collection controlled by the user, allows perceived and
real privacy concerns to be alleviated.
The core concept of local processing (on the user mo-
bile device) of health data for anonymized submission
requires that individual components of a data submission
have an associated quasi-identifier score (QIS). To avoid
the QIS exceeding a privacy threshold, data components
can be modified to be more generalized (see Section 5)
such a s for e xa mp le a s ubmiss i o n includ i ng t he c o u nt y o f
submission rather than postcode, and the QIS would re-
flect the increased generality. The approach also takes
into account the case where multiple quasi-identifiers are
submitted together as such a group of quasi-identifiers
would have a combined QIS value that is assessed
against privacy thresholds. The four core data compo-
nents in determining the combined QIS are (i) Measures,
(ii) Location, (iii) Temporal; and (iv) Demographic and
are described below
Measures are aggregate or calculated values that refer
to a specific value to be collected. Exa mple s are listed i n
Section 2.2. Location refers to the specific location a
measure occurred, Temporal refers to the period of time
in which a measure occurred and Demographic refers to
the other characteristics of an individual.
Assuming that the data submitted is aggregated across
a relatively lo ng temporal perio d, and not submitted with
exact physical location data, the only likely source of
re-identification might be the individual’s particular de-
The types of demographic data needed for the public
health data capture system, such as age or age range,
gender, major ethnicities, city or zip/postcode are also
generally non-identifying based on population distribu-
tions. The p opulat ion demo gra phics o f regions a nd coun-
tries are already collected and known in many cases such
as where national census data is collected, and in some
cases are known for specific activities that may be used
in measures, such as cycl ing-based activity . As such,
the probability of a combination of demographics can be
calculated and compared against a privacy threshold set-
ting. Such a formula for the QIS, DQIS, is below where
the λs are the indi vidual demographic details.
DQIS = 1 - Pr(λ1, λ2, λ3, …, λn)
We consider how non-identifying such demographics
might be in the foll owing evaluative section.
5. Privacy Analysis
To demonstrate the operation of this approach we eva-
luate an example data submission for the New South
Wales geographical area based on real distributions data.
This area has a population of 6,917,658 as of last census.
Using the Australian Bureau of Statistics census popula-
tion statistics  we generated a random data set based
on the relative size of the demographics, specifically
looking at gender, age bracket and local government area.
Additionally, to create plausible activity measures we
then generated activity averages and cycling participation
R. STEELE, A. CLARKE
Copyright © 2013 SciRes. CN
based on previous research .
Assessing our local processing approach we generated
the data set out to a specific number of participant sub-
mission numbers: 10000, 50000, 100000, 200000 and
400000. We then tallied the number of individuals that
had a k value under the threshold of 20, 10 and 5. Having
a small k value for a specific demographic category is
undesirable, as it can allow for potential re-identification
or inference-based attacks to be used against the data set.
As can be seen in Figure 2, at 10,000 submissions,
there were high numbers of individual submissions that
ha d l ow k values with 4782 submissions having a k value
lower than 20 and 929 having a k value lower than 5. In
practice this would be extremely problematic in ensuring
anonymity and privacy of data submissions. As for ex-
ample, if additional knowledge that an individual parti-
cipates in the population data submission is available, it
may be enough to perform re -identi fication of so me indi-
viduals. As the data submissions are increased to 400,000
these risks di minish but there is sti ll a reasonable cha nce
of re-identification even at significant data collection
levels of 400,000.
To improve this result we implemented our demographic
formula and set a reasonably conservative threshold value
for DQIS. As local government area was the optional val-
ue in this submission that was adjusted, rather than just
withholding the submission. If a DQIS value for an indi-
vidual was over the threshold based on known population
demographics, locale government area details were ex-
cluded fr om t he submis s i on.
As demonstrated in Figure 3 this resulted in a de-
crease in the number of unique submissions that had low
k values. This differentiation increased as the number of
data submissions increased with the adjusted submission
approach reaching a safe level much sooner at ~200,000
and comprising as low as 1.6% of the submissions below
the k threshold at the 100,000 submission level compared
to 4.1% in the unadjusted data set.
The threshold at the local device level could of course
be adjusted either higher or lower based on the expected
submission numbers. However, it performed quite res-
pectably at the initial level with a significantly lower
Figure 2 . Demographic k value wit hout local processi ng.
level of risk at the 10,000 submission level and close to
no statistical risk at the 400,000 level which represents
5.8% of the area total population.
The limitation of this local processing approach as
compared to a trusted server approach that performs
k-anonymity is that the number of other submissions
cannot be known with certainty by the local device. As
such, the privacy threshold is set at a conservative value
to preserve privacy. However this means that when there
are high levels of submissions more records are adjusted
than was required. This relationship is displayed in Fig-
ure 4 where for 10,000 records the percentage of records
adjusted was less than the low k value percentage and the
miss rating was extremely low. This diverges as the
number of data submissions increases, since the adjust-
ment level remains fairly constant at around 39.5% of
data submissions for the example data set. In this case
due to the high number of local government areas (153)
with a significant proportion having extremely low pop-
ulations, the adjustment rate was quite h igh.
Overall, this wouldn’t pose a serious problem, as the
priority of the demographic detail is controlled by the
data requestor and trade-offs are to be expected for in-
creased detail in other sections.
In summary, for the example data set the local
processing aggregate data approach performed favoura-
bly for the defined public health and privacy require-
Figure. 3. De mographic k value w ith local processing.
Figure 4. Adjusted submission compared to low k value
R. STEELE, A. CLARKE
Copyright © 2013 SciRes. CN
We have described that the current state of IoT in rela-
tion to commercial health and fitness sensors is well
matched to capturing data relevant to numerous public
health measures. We have described a novel IoT-based
public health information system architecture that allo ws
the completely new capabilities of both anonymous pub-
lic health data collection and anonymous public health
inter ventio n. W e have described its privacy and anonym-
ity mechanisms and provided a privacy evaluation.
 P. Klasnja and W. Pratt, “Methodological Review:
Healthcare in the Pocket: Mapping the Space of Mo-
bile-Phone Health Interventions,” Journal of Biomedical
Informatics, Vol. 45, No. 1, 2012, pp.
 A. Clarke and R. Steele, “Health Participatory Sensing
Networks,” Mobile Information Systems, 2013 (in press).
 K. Sampigethaya and R. Poovendran., "A Survey on Mix
Networks and Their Secure Applications," Proceedings of
the IEEE, Vol. 94, No. 12, 2006, pp.
 S. Mauw, J. H. S. Versch uren and E. P. Vink, "A Forma-
lization of Anonymity and Onion Routing," in (Samarati,
P., Ryan, P., Gollmann, D., and Molva, R., 'eds.'): Com-
put er Secu rit y – Esorics 2004, Springer Berlin Heidelberg,
2004, pp . 109-124.
 M. Swan, “Sen sor Mania! The Internet of Things, Weara-
ble Computing, Objective Metrics, and the Quantified
Self 2.0,” Journal of Sensor and Actuator Networks, Vol.
1, No. 3, 2012, pp. 217 -253. doi:10.3390/jsan1030217
 R. Steele, "Social Media, Mobile Devices and Sensors:
Categorizing New Techniques for Health Communica-
tion." Proceedings of the Fifth International Conference
on Sensing Technology (ICST), 2011, pp. 187-192.
 R. Steele, A. Lo, C. Secombe and Y. K. Wong, “Elderly
Persons’ Perception and Acceptance of using Wireless
Sensor Networks to Assist Healthcare,” International
Journal of Medical Informatics, Vol. 78, No. 12, 2009, pp.
 Jawbone, “UP™ by Jawbone® with MotionX® Tech-
nology Empowers You to Live a Healthier Life,” 2011
[accessed 2013 April 20]; Available from:
 Mobi Health News, “Google Adds Activity Tracking to
Android App,” 2012, [accessed 18th Jun 2013]; Online:
 Sano Intelligence [Online], Accessed 10th June 2013,
 B. Network, “Riderlog,” 2011; Available from:
 C. Outram, C. Ratti and A. Biderman, “The Copenhagen
Wheel: An Innovative Electric Bicycle System that Har-
nesses the Power of Real-time Information and Crowd
Sourcing,” in EVER Monaco International Exhibition &
Conference, 2010, pp. 8.
 D. E. Warburton, W. C. W. Nicol and S. S. Bredin.
"Health Benefits of Physical Activity: The Evi-
dence," Canadian Medical Association Journal, Vol. 174,
No. 6, 2006, pp. 801-809. doi:10.1503/cmaj.051351
 R. Steele, “An Overview of the State of the Art of Auto-
mated Capture of Dietary Intake Information,” Critical
Reviews in Food Science and Nutrition, 2013.
 AIHW, “Bi omedical Co mponen t of the Australi an Health
Survey: Public Health Objectives,” 2011, [Online].
 P. Kalnis and G. Ghinita, "Spatial K-Anonymity", in (Liu,
L., and Özsu, M.T., 'eds.'): Encyclopedia of Database
Systems, Springer US, 2009, pp. 2714-2714.
 A. Clarke and R. Steele, "Summarized Data to Achieve
Population-Wide Anonymized Wellness Measures," En-
gineering in Medicine and Biology Society (EMBC), 2012
Annual International Conference of the IEEE, 2012, pp.
 A. Clarke an d R. S teele, “S ecure and Rel iable Distributed
Health Records: Achieving Query Assurance across Re-
positories of Encrypted Health Data,” in System Science
(HICSS), 2012 45th Hawaii International Conference on,
2012, pp . 3021-3029.
 Austroads, "Australian Cycling Participation "2011,
 Australian Bureau of Statistics, "Census Community Pro-
files Greater Sydney", 2011,