Augmenting Cloud Connectivity with Opportunistic Networks for Rural Remote Patient Monitoring

Current remote patient monitoring (RPM) systems are fully reliant on the Internet. However, complete reliance on Internet connectivity is impractical in low resource and remote environments where modern infrastructure is often lacking, power outages are frequent, and/or network connectivity is sparse (e.g. rural communities, mountainous regions of Appalachia, American Indian reservations, developing countries, and natural disaster situations). This paper proposes supplementing intermittent Internet with opportunistic communication to leverage the social behaviors of patients, caregivers, and society members to facilitate out-of-range monitoring of patients via Bluetooth 5 during intermittent network connectivity. The architecture is evaluated using U.S. Census Bureau, the National Cancer Institute's, and IPUMS ATUS sample data for Owingsville, KY, and is compared against a delay tolerant RPM case that is completely disconnected from the Internet. The findings show that with only 0.30 rural population participation, the architecture can deliver 0.94 of non-emergency medical information with at least half of the information having a latency of 5 hours. In addition, the paper provides insights on how supplemented networks can be used in real-world rural RPM (RRPM) systems for different domain applications.

READ FULL TEXT VIEW PDF
POST COMMENT

Comments

There are no comments yet.

Authors

page 1

page 5

01/07/2020

Utilizing Opportunistic Social Networks for Remote Patient Monitoring in Rural Areas

The use of Internet connectivity for remote patient monitoring is often ...
01/31/2018

A Delay-Tolerant Payment Scheme Based on the Ethereum Blockchain

Banking as an essential service can be hard to access in remote, rural r...
11/18/2019

The geopolitics behind the routes data travels: a case study of Iran

The global expansion of the Internet has brought many challenges to geop...
06/11/2019

Chocolatine: Outage Detection for Internet Background Radiation

The Internet is a complex ecosystem composed of thousands of Autonomous ...
06/08/2020

Exergames for telerehabilitation

Recent advancements in technology have improved the connectivity between...
07/24/2020

COVID-19 Remote Patient Monitoring: Social Impact of AI

A primary indicator of success in the fight against COVID-19 is avoiding...
06/17/2021

Remote Health-Monitoring of First Responders over TETRA Links

In this paper, we present a system for remote health-monitoring of first...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

I Introduction

The ubiquity of mobile devices and rapid improvement in wireless body sensors has revolutionized the field of healthcare. Through mHealth solutions, practitioners can remotely monitor and assist with patients’ disease management in real time or asynchronously. This has improved the timeliness of clinical decision making, decreased the length of hospital stays, and reduced mortality rates [1, 2]. Although many patients have benefited from mHealth solutions, and national efforts are underway to accelerate broadband deployment in under-served areas of the US [3], rural patients may not benefit to the same extent as their non-rural counterparts due to geographical and financial barriers that result in limited or nonexistent access to broadband connectivity [4]. Additionally, chronic disease is approximately 20% more prevalent in rural areas than other areas [5].

A major limitation of mHealth solutions in rural areas is financial burden. The cost associated with deploying infrastructure that facilitates complete connectivity is often too expensive for many rural cities to afford without significant financial assistance [3]. For example, the KentuckyWired project in Appalachian Kentucky costs at least $324 million, with taxpayers having to pay $1.5 billion over 30 years111Lexington Herald Leader, “Cost overruns in troubled Kentucky broadband project near $100 million, audit finds”, 9/27/18, https://www.kentucky.com/latest-news/article219058225.html. A promising solution lies in the synergistic use of the aforementioned solutions. Specifically, a hybrid network architecture that leverages human mobility and minimal Internet infrastructure for disseminating patient health information.

However, the use of human mobility for remote patient monitoring (RPM) can be cumbersome due to its inherent characteristics. Unlike other networks in which the entire population is able to actively participate, patient monitoring networks typically consist of a small percentage of the population, patients, caregivers, and healthcare providers. Consequently, there are usually fewer participants in healthcare monitoring networks than in other social networks. Furthermore, RPM systems consist of a vast diversity of domain applications with different network performance requirements. For example, a pacemaker monitoring application has different network requirements from a distress measurement application. Therefore, it is imperative to understand the limitations posed by the domain application in order to design the right network for it. Hence, the authors propose a novel hybrid architecture that leverages minimal Internet infrastructure along with node mobility for the dispersal of patient health information (PHI) that is not time-critical.

The contributions of this work are it: (1) describes a network model for rural remote patient monitoring (RRPM); and (2) derives a simple mathematical description for remote monitoring in rural communities which can be used by decision makers to decide an optimal network topology using the available infrastructure; and (3) provides insights on how the number of rural patients and participating population relates to network performance. The remainder of the paper is structured as follows: Section II discusses the related work; Section III describes the network and its key members; Section IV introduces the mathematical model; Section V offers the results obtained from the model simulation; Section VI discusses the limitations of the model along with future work; and finally, Section VII concludes the paper.

Ii Related Work

Various studies have proposed different means of providing connectivity for RRPM. One such solution entails the use of mobile packet data networks to transmit PHI remotely [6]. However, the use of mobile packet data networks to transmit PHI remotely requires constant phone coverage which is often not feasible in many rural areas222Vermont Department of Public Service, “Mobile Wireless in Vermont”, 01/15/19, https://publicservice.vermont.gov/sites/dps/files/documents/Connectivity/BroadbandReports/2019/Mobile%20Wireless%20Report.pdf. Similarly, delay tolerant schemes for transmitting PHI via vehicular ad hoc networks have been proposed [7, 8]. However, the schemes solely utilize vehicular communications in transmitting messages to medical entities. Other solutions attempt to ration the low bandwidth or ignore the problem of limited Internet connectivity [9, 10, 11]. Unlike the aforementioned works, this paper is the first to evaluate the performance of a network that opportunistically leverages the natural mobility of nodes and intermittent cloud connectivity for RRPM.

Iii Hybrid Network Model

Iii-a Entities in a rural remote patient monitoring system

The proposed hybrid network consists of a DTN that supplements an intermittent Internet-connected network. The DTN consists of device-to-device (D2D) communication via Bluetooth 5 (or other D2D wireless technology) in order to facilitate out-of-range monitoring of patients. Using a hybrid network, the system is able to harness the mobility of members of the rural population and maximize data delivery without exceeding an understood latency. Although this work solely focuses on a RRPM network design, the model described can be extended to represent any other hybrid network, including heterogeneous inter-band networks and intra-band networks. The specifics of the evaluated hybrid network are discussed in Section III-B. The following sections describes the essential entities in the network.

Iii-A1 Patient

The primary data generator in a RRPM system is the patient. Objective patient data can be obtained from biosensors/wearables that form a body sensor network and transmitted asynchronously from a patient to a medical entity. Conversely, subjective data, such as surveys and assessment forms can also be collected and transmitted to the patient’s corresponding medical entity.

Iii-A2 Caregiver

In managing chronic illnesses, most patients have at least one caregiver that actively participates in the patient’s care [12]

. Unlike the patient, they are often more mobile yet they remain close enough to the patient to attend to their needs. Hence, the proposed model classifies caregivers as active transport agents because they are the nodes that have both high access to patients and high mobility. Additionally, caregivers are likely to encounter points of interests (POIs) and intermediary nodes when running errands for the patient.

Iii-A3 Intermediary Network

The intermediary network consists of intermediary nodes that are members of the society and already participate in other networks such as social networks, transportation networks, mail-delivery systems, trash collection, and/or school bus transportation systems. The intermediary network can be leveraged to opportunistically collect and deliver data without significantly increasing the costs of deploying Internet in rural areas. In relation to the RRPM network, they are classified as primary transport agents that frequently encounter POIs and caregiver nodes. Intermediary nodes move through the network differently depending on their employment state and they are more likely to have intermittent Internet connectivity.

Iii-A4 Clinical Staff

The clinical staff network consists of a set of nodes whose primary employment location is with the medical entity. These nodes are mobile and have the highest probability of encountering the destination.

Iii-A5 Points of Interests

POIs are stationary nodes or locations where transport agents typically congregate. Some examples include grocery markets, post offices, banks, places of worship, places of work, and city halls. Internet access points can be added to POIs based on budget constraints.

Iii-A6 Destination

The patient’s associated medical entity is the primary destination in this model. Unlike other nodes, the medical entity is stationary and is less likely to have random encounters with the intermediary network. Nevertheless, it is fully connected to the Internet and able to receive messages from any node through the Internet or D2D connection.

Iii-B Architecture

The proposed network maximizes the potential of DTN and sparse Internet connectivity by utilizing the current or future connectivity of relay nodes to transmit information to destinations. The hybrid network depends on the strength of the intermediary network for message delivery as the source nodes and their caregivers typically do not form a large enough density to facilitate efficient data delivery solely by opportunistic encounters. Additionally, some of the nodes in the intermediary network have intermittent Internet connectivity that is opportunistically leveraged to transmit medical data, increasing the delivery probability and decreasing latency.

Figure 1 is a depiction of a hybrid network where intermittent cloud connectivity is supplemented with delay tolerant opportunistic communication. Time represents a discrete time and represents one or many instances of future times and/or durations such that . At time , the patient’s mobile device aggregated body sensor information from local sensors on the patients body along with survey data inputted by the patient. In addition, the patient’s mobile device attempted to communicate with the Clinical Staff, but there was neither Internet available nor was there a direct line-of-sight connection between them. For the rest of the example, mobile devices are associated with their respective owners, and will be referred to by their owners name, such as “Patient”. Also at time , the Patient was in D2D range of Caregiver and forwarded their encrypted aggregated medical information intended for the Medical Entity, to Caregiver. During , the Caregiver encountered a member of the intermediary network and forwarded the Patient’s encrypted medical information. Since no network participant encountered the Doctor while and no participating user carrying the Patient’s message had an Internet connection, the data path to deliver the Patient’s data to the Clinical Staff has not been established.

Fig. 1: Hybrid Network for Rural Remote Patient Monitoring

At a new time period, , all users were mobile resulting in new encounters in the network; the Caregiver to the medical entity, and a member of the intermediary network to a Clinical Staff. The Patient lost their D2D connection to their Caregiver and the D2D connection between the Caregiver and the intermediary node is no longer available. In addition, the Medical Entity still has an Internet connection and the Clinical Staff has gained Internet access. A new D2D connection is formed between an intermediary node and the Clinical Staff, who has an Internet connection. Another D2D connection is also formed between the Caregiver and Medical Entity. Hence, the Medical Entity is able to receive the Patient’s data from the Internet or D2D; whichever occurs first.

Iv Modeling Rural Remote Patient Monitoring

In order to understand the viability and feasibility of the network, a simple mathematical model is introduced that depicts some real-world characteristics of the RRPM problem, including the high intermediate node to source-node ratio, limited number of data producers, prevalent presence of fully mobile caregivers and limited number of POIs.

Iv-a Model description

Consider a network in which represents a set of nodes that are randomly distributed in a square grid divided into cells, where an individual node is represented by . Let , where subsets are defined as: - patients, - clinical staff, - caregivers, - intermediary nodes, and - destinations, such that:

(1)

Let the set represent employed and unemployed intermediary nodes respectively. Furthermore, assume each node has a connectivity parameter, that defines its current connectivity status where:

Internet available (2a)
D2D available (2b)

In addition, assume that for the set that , for set that , and for there exists a probability, , for which and for which . The probability, , of a node having Internet connectivity at a certain time is determined by the rural community’s broadband access rate.

Let an individual patient be represented by and the set of messages for an individual patient be:

(3)

where message number is generated at time and is transmitted to at a time . For each message , there exists a set of nodes, , that have a copy of message and a set of nodes that do not have the message. At each distinct time, = [], encounters occur between nodes and through those encounters, messages in are transmitted. Once a node in encounters a node, , in , the corresponding message, , is transmitted and becomes a member of set .

After obtains the message and is added to the node set, if = and , then nothing else happens. However, if = or if , the message is considered to be delivered and the difference between the start time and the time, , at which it occurs is the delivery latency for that message, . At each consecutive time step, more encounters occur. Finally, when , the delivery probability can be calculated as the number of messages in transmitted to .

(4)

Additionally, the upper-bound delivery latency for all delivered messages is defined as the message with the largest or:

(5)

Iv-B Mobility and transmission

The mobility of nodes in the network is described by

discrete time Markov chains with a finite number of states

[13, 14]. For simplicity, the following states are used: home, work, and POI. Individual home and work locations are assigned to each node and POIs can be randomly selected from the set, , of POIs during each transition. The subset {D,P} are considered stationary nodes and do not have a transition matrix associated with them. Each mobile node in subset has a unique transition matrix for each time period, . Where, each period starts at and consists of consecutive time steps. For example, employed nodes such as and are preferentially attached to and are stationary at work locations, which consists of POIs in the grid during the work period (e.g. 9:30am - 4:30pm). Hence, and nodes are mobile between home and work. The transition probabilities can be derived from an observation of the rural community in question. The model assumes that contact occurs when two nodes with the same radio are within transmission range of each other where, the transmission range is assumed to be circular. Messages are also assumed to be small enough to be successfully transmitted within each encounter and uniformly sized.

V Evaluation

V-a Modeling a real rural community

In evaluating the feasibility and viability of the model, data was obtained from the Federal Communications Commission and the US Census Bureau regarding Owingsville, KY (Bath County)[15]. Owingsville, KY was chosen because it is a rural city with health and connectivity issues while having a moderate technology adoption rate.

Time (Period)

Initial Probability Vector and Transition Matrices (

, , )
Node Classification: {} Node Classification: {}
19:00 - 06:30 (1) (0.85, 0, 0.015)  (0.70, 0.079, 0.22) 
06:30 - 09:30 (2) (0.93, 0, 0.070)  (0.71, 0.16, 0.13)
09:30-16:30 (3) (0.76, 0, 0.24) (0.50, 0.33, 0.13)
16:30-19:00 (4) (0.77, 0, 0.23) (0.48, 0.20, 0.32)
TABLE I: Transition Matrices Derived from ATUS Data.

V-B States and transitions

Based on a 2017 IPUMS ATUS sample of non-metropolitan households in the US, 303 routine activities were obtained, along with corresponding start and stop times, and classified into three states: Home, Work and POI [16] as described in Section IV-B

. In addition, information from IPUMS ATUS was obtained for the number of individuals in each state for 30 minute intervals, and four (4) periods were defined based on the number of people in each state. The four (4) periods defined were: 1) 19:00 - 06:30, 2) 06:30 - 09:30, 3) 9:30 - 16:30, 4) 16:30 - 19:00. Consequently, the transition matrix was estimated for each period by obtaining the transition matrix for each individual, and aggregating it over each period. The resulting matrices and periods are given in Table

I.

Parameter Value Source
Simulation seeds 0:1:99
Simulation duration 24 hours
Adult Population of Owingsville 400 [15]
Area of Owingsville 2.409 sqmi [15]
Number of Cells [15]
Cell size 10 ft 10ft [15]
Number of patients () 2:2:10 [17]
Number of caregivers () 2:2:10 [18]
Number of destinations () 1 [18]
Ratio of population involved in intermediary network () 0.1:0.1:1
Ratio of Internet connected intermediary nodes 0.2
Number of POIs () 25 Map
Number of Clinical Staff () [19]
Periods 1 to 4 [16]
Data generation rate 1 message per 24 hours Markey Cancer Center
Ratio of employed nodes 0.935 [20]
Transmission range (based on Bluetooth 5) [21]
TABLE II: Parameters used in Simulation

V-C Simulation setup

To understand how the proposed hybrid network, described in Section IV-A, and a DTN, completely disconnected from the cloud, could be used for RRPM, distress information communication from cancer patients (source nodes) to their respective healthcare providers (destination nodes) was used as a domain example and a simulation environment was created in Python333Code available at https://github.com/netreconlab/globecom2019. For simplicity and understanding of the capabilities of the model, an ideal version of the Epidemic routing protocol was used that did not limit the abilities to broadcast or send information (i.e. buffer limitations). For all simulations, patients were mobile, were the only generators of data as mentioned in Section III-A1, and never had Internet connectivity. The number of cancer patients was varied from 2, which is the estimated number of lung cancer patients in Owingsville to 10, which is the estimated number of all cancer patients in Owingsville [17]. One message was generated per patient at the beginning of the simulation for 100 seeds as shown in Table II. To determine the effect of message-generation time on delivery probability and delivery latency in the DTN and the hybrid networks, the initial message generation period was varied between the four (4) periods defined in Table I for 100 seeds. Finally, the number of intermediary nodes were varied in the network for 100 seeds to determine its effect on delivery latency and delivery probability. Table II describes the rest of the parameters used in the simulation along with their sources for their values.

V-D Results

Figures (a)a and (b)b

show the effect of the period at which a message is generated on the delivery latency and ratio of messages delivered. At each period, there is an average of 0.9 increase in delivery probability associated with using the hybrid network compared to the DTN. Additionally, a matched-pairs t-test reveals that there is a statistically significant decrease in the delivery latency for hybrid network compared to the DTN when messages are generated in periods 3 and 4.

Similarly, regardless of the number of patients in the network, there is a statistically significant increase in delivery probabilities and a decrease in, or unchanged delivery latency between the hybrid and DTN (Figures (c)c-(d)d). Figures (e)e and (f)f exhibit the effects of the number of relay nodes in the network on the delivery latency and ratio of messages delivered. As more intermediary nodes are added to the network, the delivery probability approaches . However, there is a saturation point, 0.5, at which an increase in population participation does not influence the delivery probability. Increasing population participation does continue to reduce delivery latency as shown in Figure-(d)d, after the saturation point, the latency associated with the hybrid network decreases more rapidly than that of the DTN.

(a) Mean delivery
(b) Median delay
(c) Mean delivery
(d) Median delay
(e) Mean delivery
(f) Median delay
Fig. 2: (a) and (b) vary the period at which a message is generated with .20 population participation in the network and 10 patients generating messages. (c) and (d) vary the amount of patients with .20 of the population participating in the network, each patient generating messages at the start of the second period. (e) and (f) vary population participation in the network with 10 patients generating a message at the start of the second period. * signifies a statistically significant difference at .05 significance, error bars represent SEM.

Vi Discussion and Future Work

Upon evaluation, the proposed hybrid network consistently shows significantly higher delivery rates at lower or comparable delivery latencies. Based on the results in Section V-D, one can infer that the hybrid network may be most suitable for certain domain applications requiring a lower delivery latency. For example, a RRPM application that requires a delivery rate of .95 and a delivery latency with an upper bound of 3 hours may be most suitable for the hybrid network. Conversely, an RRPM application that requires a delivery rate of .95 and a delivery latency with an upper bound of 2 minutes will require Internet connectivity for all devices and systems in the network.

While the proposed solution does provide an innovative means of transmitting patient data, it is not without limitations. One such limitation is need for population participation to increase the amount of intermediary nodes. As a result of this limitation, the authors hope to explore a means of modeling node incentivization that represents human social behaviour. In addition, the benefits of the hybrid model occur only in areas where there is sporadic Internet connectivity and assumes that healthcare providers have full connectivity. If Internet connectivity is never available, the network will fall back to the DTN scenario.

Another limitation of this work is that it only applies to delay tolerant situations where information between patient and provider is not emergency related or time-critical. However, alternative means of improving network performance can be explored along with using mobile devices at the edge to respond to time-critical messages/data. In addition, this work only considers the dissemination of small data packets using the theoretical transmission range of Bluetooth 5. Future work will consider the behaviour of this network model when the size of data is perturbed with real-world transmission ranges. Also, this paper does not include the effects of privacy, security, ethics, and data use agreements as they are beyond the scope of this paper. However, security measures can be implemented as described in previous work [22, 23]. Lastly, the proposed architecture does not discuss optimal DTN routing protocols to use for D2D communication. Future studies will focus on modeling and harnessing pertinent characteristics of rural communities in order to create adaptive routing protocols for such areas. The aforementioned future work will culminate into an applied evaluation and deployment of a RRPM system for lung cancer patients in Appalachian Kentucky along with understanding the proposed architectures capabilities.

Vii Conclusion

People living in rural America suffer from life threatening illnesses (e.g. lung cancer) while experiencing the lack of access to timely and quality care due to insufficient healthcare resources and Internet connectivity. The authors propose a novel architecture that supplements intermittent Internet coverage by transmitting patients’ health data opportunistically until it reaches healthcare providers. The simulation results, using real-world data from Owingsville, KY, a small rural Appalachian city, have demonstrated that the proposed model is feasible and can provide a timely and reliable communication to remotely link rural patients with their providers; resulting in better quality of care. The authors will continue developing the model with stakeholders in Appalachian Kentucky communities and testing it as part of a large, public-private partnership supported project, the Linking & Amplifying User-Centered Networks through Connected Health (LAUNCH) initiative444LAUNCH is a partnership between the Federal Communications Commission’s Connect2Health Task Force; the National Cancer Institute; the University of Kentucky Markey Cancer Center; the University of California, San Diego Design Lab; and Amgen. More info here: http://launchhealth.org.

References

  • [1] A. Niksch, B. Rothman, N. Hodge, and M. Ranney, “The value of remote patient monitoring (rpm) physicians’ perspectives,” Healthcare Information and Management Systems Society (HIMSS), 2014.
  • [2] E. Moy, “Leading causes of death in nonmetropolitan and metropolitan areas—united states, 1999–2014,” MMWR. Surveillance Summaries, vol. 66, 2017.
  • [3] USDA, “A case for rural broadband,” United States Department of Agriculture, Tech. Rep., 2019. [Online]. Available: https://www.usda.gov/sites/default/files/documents/case-for-rural-broadband.pdf
  • [4] S. J. Henley, R. N. Anderson, C. C. Thomas, G. M. Massetti, B. Peaker, and L. C. Richardson, “Invasive cancer incidence, 2004–2013, and deaths, 2006–2015, in nonmetropolitan and metropolitan counties—united states,” MMWR Surveillance Summaries, vol. 66, no. 14, p. 1, 2017.
  • [5] CDC, “New cdc report shows deaths from cancer higher in rural america,” Centers for Disease Control and Prevention, Jul 2017. [Online]. Available: https://www.cdc.gov/media/releases/2017/p0706-rural-cancer-deaths.html
  • [6] A. Bolaji, “Simulation of a real-time mobile health monitoring system model for hypertensive patient in rural nigeria,” Afr J Comp ICT, vol. 7, pp. 95–100, 2014.
  • [7] M. Barua, X. Liang, R. Lu, and X. S. Shen, “Rcare: Extending secure health care to rural area using vanets,” Mobile Networks and Applications, vol. 19, no. 3, pp. 318–330, 2014.
  • [8] M. J. Murillo and M. Aukin, “Application of wireless sensor nodes to a delay-tolerant health and environmental data communication system in remote communities,” in 2011 IEEE Global Humanitarian Technology Conference.   IEEE, 2011, pp. 383–392.
  • [9] D. Duggan, J. A. Muh, L. Ndongo, J. Yao, and L. Yu, “Mobile device administration for secure and manageable health data collection in under-resourced areas,” in Proceedings of the 7th Annual Symposium on Computing for Development, ser. ACM DEV ’16.   New York, NY, USA: ACM, 2016, pp. 2:1–2:10. [Online]. Available: http://doi.acm.org.ezproxy.uky.edu/10.1145/3001913.3001927
  • [10] M. Pourhomayoun, N. Alshurafa, F. Dabiri, K. Yadav, C. Sideris, L. L. Tseng, H. Ghasemzadeh, A. Nyamathi, and M. Sarrafzadeh, “A robust remote health monitoring and data processing system for rural area with limited internet access,” in Proceedings of the 11th EAI International Conference on Body Area Networks.   ICST (Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering), 2016, pp. 26–32.
  • [11] K. Adlassnig et al., “Rural telemedicine networks using store-and-forward voice-over-ip,” in Medical Informatics in a United and Healthy Europe: Proceedings of MIE 2009, the XXII International Congress of the European Federation for Medical Informatics, vol. 150.   IOS Press, 2009, p. 448.
  • [12] R. Hughes, Patient safety and quality: An evidence-based handbook for nurses.   Agency for Healthcare Research and Quality Rockville, MD, 2008, vol. 3.
  • [13] J. Geweke, R. C. Marshall, and G. A. Zarkin, “Mobility indices in continuous time markov chains,” Econometrica: Journal of the Econometric Society, pp. 1407–1423, 1986.
  • [14] A. Picu, T. Spyropoulos, and T. Hossmann, “An analysis of the information spreading delay in heterogeneous mobility dtns,” in 2012 IEEE International Symposium on a World of Wireless, Mobile and Multimedia Networks (WoWMoM).   IEEE, 2012, pp. 1–10.
  • [15] “U.s. census bureau quickfacts: Bath county, kentucky,” United States Census Bureau. [Online]. Available: https://www.census.gov/quickfacts/bathcountykentucky
  • [16] S. L. Hofferth, F. M. Flood, and M. M. Sobek, “American time use survey data extract builder: Version 2.7 [dataset],” 2018. [Online]. Available: https://doi.org/10.18128/D060.V2.7
  • [17] NIH-NCI and CDC, “Incidence rate report for kentucky by county,” National Institute of Health - National Cancer Institute, Centers for Disease Control and Prevention. [Online]. Available: https://www.statecancerprofiles.cancer.gov/incidencerates/index.php?stateFIPS=21&cancer=001&race=00&sex=0&age=001&type=incd&sortVariableName=rate&sortOrder=default#results
  • [18] L. L. Berry, S. M. Dalwadi, and J. O. Jacobson, “Supporting the supporters: What family caregivers need to care for a loved one with cancer,” Journal of oncology practice, vol. 13, no. 1, pp. 35–41, 2016.
  • [19] “Discover u.s. government information: Title 42 - public health,” May 2014. [Online]. Available: https://www.govinfo.gov/content/pkg/CFR-2017-title42-vol5/xml/CFR-2017-title42-vol5-sec491-8.xml
  • [20] “Unemployment rate in bath county, ky [kybath1urn],” U.S. Bureau of Labor Statistics retrieved from FRED, Federal Reserve Bank of St. Louis. [Online]. Available: https://fred.stlouisfed.org/series/KYBATH1URN
  • [21] M. Woolley, “Bluetooth 5: Go faster, go further.” Bluetooth SIG, Tech. Rep., Apr 2019. [Online]. Available: https://www.bluetooth.com/bluetooth-resources/bluetooth-5-go-faster-go-further/
  • [22] C. E. Baker, A. Starke, T. G. Hill-Jarrett, and J. McNair, “In vivo evaluation of the secure opportunistic schemes middleware using a delay tolerant social network,” in Distributed Computing Systems (ICDCS), 2017 IEEE 37th International Conference on.   IEEE, 2017, pp. 2537–2542.
  • [23] E. Max-Onakpoya, A. Jacobs, and C. E. Baker, “An opportunistic mhealth architecture for remote patient monitoring,” in Proceedings of the 20th International Workshop on Mobile Computing Systems and Applications.   ACM, 2019, pp. 169–169.