Dimagi/CIDRZ Group


Intent

We developed a software application that would facilitate the use of mobile device to transport images taken by nurses at health care centers (point-of-care) in rural Zambia to a server in Lusaka where doctors can access the images in order to provide expert advice on cervical lesions. We plan to use this application to facilitate a scalable national cervical cancer screening program.


 

Background

Cervical cancer is a leading cause of death from cancer among women in resource-poor countries, affecting women at a time of life when they are critical to social and economic stability [1].  Its rise in prevalence over the last decade is a direct consequence of the spread of HIV.  Transmission of the human papilloma viruses that cause cervical cancer is facilitated by the presence of HIV infection.  In addition, progression of cervical dysplasia is accelerated by immunosuppression that results from an advancing HIV infection.  Cervical cancer is one of the few cancers with a long preclinical phase.  A screening program aimed at detecting and treating precancerous lesions reduce the number of cases that require treatment.  This is the basis for annual Pap smear screening that is routine in most developed countries.  However, this approach requires established laboratories, highly-trained cytologists, and up to 3 visits per screening (Pap smear, + colposcopy, treatment), and is not feasible in resource-poor countries.  Single visit "screen-and-treat" approach using visual inspection with 4% acetic acid (VIA) and early treatment with cryotherapy have been shown to feasible and effective in reducing cervical cancer incidence and mortality [2-4].  In a study by the Harvard School of Public Health, screening of women with one-visit visual inspection at about 35 years of age would reduce the lifetime risk by 25 to 36 percent [6].  Two screenings in a lifetime would provide a relative increase in the lifetime reduction of risk of cancer of approximately 40 percent, but obviously at a higher cost.  

The Center for Infectious Disease Research in Zambia (CIDRZ) has been providing HIV prevention and treatment services since 2004.  For this project, CIDRZ has partnered with Dimagi, an organization that provides technological solutions to problems in the developing world, to design and implement a national cervical cancer screening program.  Given that diagnostic and therapeutic expertise in cervical malignancy is scarce and localized in Lusaka, a nationwide screening using the traditional approach of performing Papanicolau smear is out of the question.  The use of visual inspection with 4% acetic acid (VIA) and early treatment with cryotherapy by rural nurses has been implemented in a number of locations in Zambia, with quality assurance provided by Lusaka-based specialists.

The use of cell phones to facilitate both real-time and store-and-forward medical consultation is an established tool for delivering specialty care, most notably dermatology and wound care [6-8]. In a review of the teledermatology literature, Whited reported that the agreement rate between clinic-based dermatologists and store-and-forward teledermatologists averaged 80% but can approach almost 95% [9].  Additional reports found that cell phone cameras can provide off site surgeons with images that can effectively be used to determine surgical care [10-11].

 

Current Landscape

CIDRZ currently has deployed the cervical cancer screening system just described in 4 clinics outside of Lusaka.  Images are taken primarily for quality assurance and quality improvement purposes.  Real-time decision support where indeterminate lesions can be reviewed by specialists is not available due to file size issues in GPRS emails.  Also for this reason, quality assurance is provided by specialists who physically travel to the clinics once a week to review the images with the nurses.  This process is not scalable given the resource constraints.


Aims and Key Deliverables

The aim is to develop an application that uses camera on device, captures image, resizes and transmits to an off-site specialist for decision support and quality assurance/quality improvement (QA/QI) purposes.  Using J2ME, we developed an application for a cervical image taken in the clinic to be transmitted to the central server, accompanied by basic patient and clinical information.  The server is located at the CIDRZ office in Lusaka.  Using LAMP, we designed a platform for server-end information management.  In this platform, the images and relevant clinical information is accessible in a real-time manner off a standard web browser via the internet.  As soon as the server receives an image with accompanying clinical information, an SMS text is sent to an on-call physician in the hospital to alert the physician of the need for decision support. 

For the next phase of the project, the central server shall provide capabilities for progress note entry by and communication between the nurses at the point-of-care and off-site specialists.  Additional features of this web portal will include a personalized inbox for nurses and doctors to manage images under their care, a simple patient tracking system to allow for rapid retrieval of a patient’s image history, and a rich markup system to annotate relevant metadata on a patient’s image for retrieval at a later time.  We will explore methods that will provide integration that would allow our application to support HL7, facilitating interface with other health information systems.

 

Insuring Sustainability

As Professor Ken Keniston pointed out, demonstrating value for ICT4D projects is key in order to insure sustainability.  Value is defined by the outcomes achieved divided by the cost of achieving those outcomes.  Interventions to address specific societal needs should undergo rigorous and robust evaluation especially when there are limited resources, whether economic or human.  The use of expensive technology should be justified by value, not by its ingenious design and demonstration of its feasibility.  It has to facilitate the most cost-effective way of achieving the desired outcomes and the delivery of the best return-on-investment.

In collaboration with the Zambia Ministry of Health, CIDRZ, Dimagi and the Harvard School of Public health, we are planning to perform cost effectiveness analysis of our technology-facilitated national cervical screening program through mathematical modeling.  This will require information including the cost of implementation of the technology, outcomes tracking (cervical cancer incidence and mortality) as well as data that are already available from other countries that are demographically similar to Zambia (e.g. natural history of cervical cancer, patterns of sexual behavior, risk factors for cervical cancer, cost of treatment of advanced cervical cancer).  It is important to note, however, that cost information differs between countries,and country-specific data are preferable as much as possible.  This analysis will allow us to calculate the cost per year of life saved, and depending on a country's cut-off for what it considers cost-effective (usually less than the per capita GDP according to the Commission on Macroeconomics and Health [12]), will determine whether the project delivers a good return-on-investment.  Such an analysis will also help us in optimizing parameters such as frequency of screening during lifetime, and the age of women who should be targeted for screening.  The traditional screen-and-treat approach using VIA, for example, was shown to be most cost-effective in 5 resource-poor countries when performed once or twice in a lifetime to women between the ages of 35 and 45 years.  Whether this finding is valid as well for our mobile device-facilitated approach remains to be seen. 

Rigorous and robust analysis of ICT4D projects, such as the one we are hoping to perform, can provide guidance for the global community by identifying health investments that are of the highest priority and have the greatest promise.  This requirement to assess sustainability should, however, in no way deter people from trying to design and implement ICT4D projects.  One can argue that it is ridiculous to talk about sustainability of projects to help others while we pour trillions of dollars to the Iraqi war "for as long as it takes" without any discussion of sustainability or value for that matter.

 

Technological Architecture
User case. The nurses will be collecting data in the form of text describing medical information as well as images of patients. The data will be collected using a camera phone that will then send the information to a server via GPRS using the local cellphone company Celtel. Doctors will access the data on the server to review the cases and give feedback back to the nurses. Since the server is at CIDRZ, doctors at CIDRZ can connect to the server directly using their local network while doctors at the hospital or in the US will be using iConnect, which provides a cheaper internet connectivity compared to Celtel.

 

Here is a diagram that summarizes this:

 



Architecture. The solution is based on J2ME midlet running on a N90 Nokia phone. The midlet will access the camera of the phone and will let the user enter text data. The data will be collected using a XForm if it's possible to easily integrate images in there. The web application will receive the data and store it into a RDBMS such as MySQL to then render it to the end user on a web browser. The user will use the web interface to anotate people's records and send that information back to the nurses as well as share it with other doctors. The server will run on either LAMP (Linux, Apache, MySQL, PHP) or something more standard for data collection such as ORBEON.


The Workflow
The nurse obtains the following clinical information from the patient, and then prepares the patient for a speculum examination.  

A cervix that is grossly abnormal does not require painting with 4% acetic acid.  Images are taken and either the patient is treated with or without real-time decision support, or referred to the hospital.  Normal looking cervices are painted with 4% acetic acid, then classified after visual inspection as either negative, indeterminate or positive.  Images of indeterminate cases are transmitted for review by off-site specialists.   

 

The images are transmitted to the server accompanied by the clinical information obtained from the patient. 

   

Once the image and clinical information are received by the server, the on-call doctor is alerted via an SMS text containing the image and a URL link to the accompanying clinical information that has just been archived in the server.  After reviewing the image and the clinical adata, the off-site doctor contacts the on-site nurse by cell phone to discuss the image and the patient.  The diagnosis and treatment plan are agreed upon, and entered into the server by the doctor through a web portal.  All the archived images and clinical information are retrieved at a later time for review by the doctors and nurses for QA/QI purposes.


Patients who receive treatment on site with either cryotherapy or antibiotics will be required to follow-up.  All the follow-up images will be sent to the central server as well for archival.  This is a very important step for QA/QI.  However, we are aware that there is a significant number of patients that are lost to follow-up.  

Internet and GPRS Providers

In Zambia right now, there are currently no hard wired internet connections.  All internet access for government and organizations alike use some variation of satellite.

Celtel is one of the larger cellphone providers in Zambia that provides GPRS and EDGE connectivity.  EDGE (2.5G) connectivity is available in the capital region of Lusaka and the densely populated Copperbelt region due north. 

For its business/data subscribers, Celtel has a variety of tiered services for bandwidth usage.  We envision for a clinic phone to use the 100 MegaByte/Month plan costing K85,000 which is about $25.00 dollars a month.  The estimated bandwidth usage for a phone in a clinic looks to be well within the boundaries of this service plan.  On average, a clinic looks to see about 400 patients a month at full capacity.  A generous file size of 150 kilobytes per image/survey datagram means that there is still enough bandwidth allowance available. 

According to Celtel, their EDGE network supports a maximum throughput of 236 kilobit/second transfer.  From a single device, sending images over this link is reasonable and the only real way to send them from a faraway clinic back to the central server.

Celtel is also looking to provide GPRS/EDGE connectivity as an enterprise ISP with unlimited bandwidth for company's connectivity needs.  As of this writing the pricing plan is still under development.  The bandwidth listed above is a concern for a central server to aggregate images being transmitted from all over the country.  One benefit of a Celtel based gateway connection is that all Celtel originating GPRS traffic will be routed internally via Celtel's network infrastructure.  This saves a significant amount of latency as a satellite transaction will no longer be necessary to route certain packets to a next hop.  Celtel therefore provides a wide area network with reasonable latency for internal network applications.  Even Celtel's internet access is dependent on satellite access.

iConnect is a satellite Internet Service Provider providing a variety of dedicated services with a maximum throughput of 512kilobit/second.  Their business ISP products have much larger total bandwidth usage limits at a substantially higher cost, ranging from $300 to $1000 dollars a month depending on the service levels desired.  In Southern Africa, these prices are the norm for high quality internet connectivity.

The server will be connected to Celtel and iConnect to take advantage of the best price for the appropriate connection.  We envision the Celtel connection to be a backup in case connectivity via iConnect is not reachable due to inclement weather or loss of power or connectivity due to other factors.  The higher cost of iConnect can be further justified as it benefits CIDRZ's other organizational and operational needs for internet connectivity.

Addressing Privacy Issue
Patients are assigned a unique identifier at the health center.  This information is stored and can only be accessed locally.  The images and clinical information are labeled with the patient's unique identifier both for transmission and for storage at the central server.  Patient names are not important during the regular quality assurance/quality improvement meetings.  For follow-ups done locally, the nurse will be able to "link" the images from the locally-stored relational database.  For patients who are referred to the hospital in Lusaka, a letter with the patient's unique identifier will be given to the patient in case the Lusaka specialists would like to access the previous images from the central server.  

Operational and Technical Risks 
The most important operational risk is the reliability of the cellular and internet infrastructure.  Power failures are likewise common in Zambia and pose another serious operational risk as internet connectivity is tied to power.  Back-up with GPRS modem is thus necessary.
Data stored in the cell phones will be stored in a computer located at each rural health clinic.  This will address potential loss of data if a cell phone is lost or damaged.
Camera and cell phone skills of the nurses and computer skills of the doctors may also be a issue for succesful implementation.  Training and evaluation of skill demonstration will need to be included in the time table for the pilot and roll-out implementation.

Quality Assurance/Quality Improvement
Quality assurance and quality improvement will be done through periodic reviews of all the images that have been transmitted.  The platform for these regular QA/QI meetings has not been decided yet. 
Metrics that will be tracked during implementation include: 

Process Metrics:
1. Number of images transmitted that were deemed suboptimal requiring the image to be re-taken and re-sent 
2. Number of images that did not go through for technical reasons
3. Time it takes for the on-call specialist to get back to the nurse with his input 

Outcome Metrics 
1. Number of images sent per nurse per week that require specialist input 
2. Number of cases where diagnosis was deemed inaccurate after review during the QA/QI process or in retrospect during follow-up

Potential Use of this Application outside Cervical Cancer Screening
This application of transmitting images and clinical data for both real-time decision support and archiving has tremendous potential to transform health care delivery in resource-poor settings.  Current applications whereby images are transmitted from cell phone to cell phone for real-time decision support or from cell phone to a computer via email for store-and-forward systems lack the functionality that would facilitate follow-up, outcomes metric tracking and quality assurance/quality improvement.  Archiving and information management in a server allows remote longitudinal care in order to leverage specialists that are usually only available in the urban centers.  Consider this hypothetical case.  A gentleman from a far-flung region is brought to Lusaka complaining of abdominal pains.  A blockage is found in his intestines, requiring an operation to remove a portion of his intestines.  He is discharged after the surgery with instructions for regular follow-up to monitor the healing of the wound from the operation.  A week later, the patient notices some redness around the edges of the wound.  But because it is a full-day trip to Lusaka, he decides against the trip and he monitors the wound himself.  A few days later, the patient dies as a result of a massive infection of the surgical wound.  This software application can therefore potentially revolutionize follow-up of patients who live in remote areas, especially if visual inspection is crucial in the clinical examination, such as in dermatology, ophthalmology and wound care.  And because the images are stored, regular peer review of cases are facilitated for QA/QI purposes.  Process and metrics tracking can also be automated within the application.

Work Plan

 

Element

Action

Time line

Requirements

Collect information about people using the system as well as the state of cellphone access in Zambia

End of March

Client prototype


 

Able to send data to server along with an image and text

 

Mid April

Server prototype

Able to receive data from a client that includes image and text and send the image to the on-call doctor for real-time decision support  

Late April

Testing and enhancements

Iteration between testing and development    

Early May


Research and Development Team

Andrés Monroy-Hernández

Leo Anthony Celi

Daniel Myung, project adviser

Gari Clifford, project adviser

 

References:

1.  Ferlay J, Bray F, Pisani P, Parkin DM. GLOBOCAN 2000: cancer incidence, mortality and prevalence worldwide, version 1.0 IARC Cancerbase no. 5, Lyon, France: IARC Press, 2001.

2.  Sankaranarayanan R, Esmy PO, Rajkumar R, Muwange R, Swaminathan R, Shanthakumari S, Fayette JM, Cherian J. Effect of visual screening on cervical cancer incidence and mortality in Tamil Nadu, India: a cluster-randomised trial. Lancet 2007;370:398-406.

3.  Denny L, Kuhn L, De Souza M, Pollack AE, Dupree W, Wright TC. Screen-and-treat approaches for cervical cancer prevention in low-resource settings: A randomized controlled trial. JAMA 2005;294(17):2173-81.

4.  Brewster WR, Hubbell FA, Largent J, Ziogas A, Lin F, Howe S, Ganiats TG, Anton-Culver H, Manetta A. Feasibility of management of high-grade cervical lesions in a single visit. JAMA 2005;294(17):2182-87.

5.  Goldie SJ, Gaffikin L, Goldhaber-Fiebert JD, Gordillo-Tobar A, Levin C, Mahe C, Wright T for the Alliance for Cervical Cancer Prevention Cost Working Group. Cost-effectiveness of cervical cancer screenin gin five developing countries. NEJM 2005;353(20):2158-68.

6.  Whited JD, Hall RP, Simel DL, Foy ME, Stechuchak KM, Drugge RJ. Reliability and accuracy of dermatologists' and clinic-based and digital image consultations. J Am Acad Dermatol. 1999;41(5, pt 1):693-702.

7.  Hofmann-Wellenhof R, Salmhofer W, Binder B, Okcu A, Kerl H, Soyer HP. Feasibility and acceptance of telemedicine for wound care in patients with chronic leg ulcers. J Telemed Telecare. 2006;12(suppl 1):15-17.

8.  Braun RP, Vecchietti JL, Thomas L; et al. Telemedical wound care using a new generation of mobile telephones. Arch Dermatol. 2005;141(2):254-258.

9.  Whited JD. Teledermatology research review. Int J Dermatol. 2006;45(3):220-229.

10.  Hsieh CH, Tsai HH, Yin JW, Chen CY, Yang JC, Jeng SF. Teleconsultation with the mobile camera-phone in digital soft-tissue injury: a feasibility study. Plast Reconstr Surg. 2004;114(7):1776-82.

11. Hsieh CH, Jeng SF, Chen CY, Yin JW, Yang JC, Tsai HH, Yeh MC. Teleconsultation with the mobile camera-phone in remote evaluation of replantation potential. J Trauma. 2005;58(6):1208-12. 

12.  World Health Organization. Macro-economics and health: investing in health for economic development: report of the Commission on Macroeconomics and Health. Geneva: World Health Organization, 2001. 

    


Appendix A. Needs Assessment

 

Who is the target population?
What demographic data do we need?
What infrastructure (physical and human) information would help you design your project/product?
What beliefs, opinions, and practices would affect utilization of your project/product?

 

Appendix B. Clinical Information accompanying Images

1.      Date and time stamp      

2.      Patient ID number  

3.      Clinic location  

4.      Practitioner ID 

5.      Age  

6.      Year of last screening  

7.      HIV status  

a.      Positive  

b.      Negative  

              Date last tested  

c.      Not tested  

8.      ARV status  

a.      On ARV  

b.      Not on ARV  

9.      CD4 count   

a.      <100  

b.      100-200  

c.      >200  

10.      Gynecologic history  

a.      Previous cervical lesion  

              Yes 

                        Infection 

                        Tumor 

                        Unclear 

                        Others. Specify: 

              No 

b.      Previous treatment for cervical lesion 

              Antibiotics 

              Cryotherapy

              Others. Specify: 

 

     11.   Nurse's Impression

            a.    Normal

            b.    Pre-cancerous

            c.    Cervical Cancer

            d.    Infection

            e.    Others

     12.    Treatment Plan (Optional)

a.        No treatment necessary. 

b.       No treatment now but needs repeat examination. Specify when: 

c.        Refer to hospital. 

d.       Cryotherapy 

e.        Antibiotics 

f.         Others. Specify: 

 

 

Appendix C. J2ME Code for Image and Data Transmission to Server (submitted separately as a zip file)

 

Appendix D. Web Interface for Image, Clinical Data, Diagnosis, Treatment Plan and Review

 

 

 

 

 

 

 

 

 

Appendix E.  Script for the Demonstration Video

the developing world, with sub-Saharan Africa being the worst-affected .  This is likely a consequence of the spread of HIV in the region.  Transmission of the human papilloma viruses that cause cervical cancer is facilitated by the presence of HIV infection.  In addition, progression is accelerated by the immunosuppression that results from HIV. 

In Zambia, cervical cancer strikes 63 women in 100,000, affecting women at a time of life when they are critical to social and economic stability.  But cervical cancer takes a long time to fully develop, and thus, is highly preventable through yearly pap smears that facilitate detection and treatment of pre-cancerous lesions.  Such screening, which is routine in countries like the US, is not possible in most countries because of the resources it requires.

Visual inspection of the cervix after painting with vinegar or VIA to detect pre-cancerous lesions has emerged as a cheaper alternative in developing countries.  Studies have demonstrated that this cheaper prevention program is effective in reducing the incidence of and the death rate from cervical cancer. 

Since 2006, through the Center for Infectious Disease Research in Zambia or CIDRZ, close to 20,000 women have undergone cervical cancer screening through VIA performed by nurses.  However, the program cannot be scaled because of constraints on health care workforce. In up to 10% of cases, it is difficult to say whether the findings on VIA are precancerous or not.  Currently, the nurses would diagnose these indeterminate cases as precancerous, treat them with cryotherapy, and arrange follow-up visits for these women.  A significant number of false-positives is thus inevitable in this set-up, leading to unnecessary follow-up visits, further constraining an already severely-constrained labor resources.  Needless to say, input from a specialist would be very helpful in these situations. 

The goal of this project is to facilitate, using information and communication technology, real-time specialist support for nurses.  This is the only way to be able to scale cervical cancer screening using VIA to the rest of the country. 

We developed a software application that allows the nurses to take photographs of the cervix, and transmits the images accompanied by relevant clinical information to a central server using GPRS.  The server then alerts a specialist-on-call that a nurse is requesting decision support.  The URL of the image and the clinical information is sent to the doctor via MMS.  The doctor then calls the nurse and reviews the case after accessing the data from the server, and a decision is made.  The final diagnosis and plan of treatment are entered into the central server by the doctor.

Archiving and information management at a central server is necessary in order to be able to track the progress of a patient, including clinical outcomes.  It is also crucial for quality assurance and quality improvement.  All the images, diagnosis and treatment plans are reviewed by the nurses and specialists regularly at a later time.  What we’d like to see is a reduction over time of not only misdiagnoses, but also of indeterminate cases requiring specialist support.  Finally, once a large database of cervical images has accumulated that are linked to both diagnoses that have been QA’d and clinical outcomes, it is likely possible to come up with artificial intelligence tools that can classify images of indeterminate cervical lesions, further reducing the need for specialist support.

Although this software application was initially developed to facilitate cervical cancer screening, it will be useful for a wide variety of scenarios in health care delivery.  For example, this software application can be a tremendous help for follow-up of patients who live in remote areas, especially if visual inspection is crucial in the clinical examination, such as in the fields of dermatology, ophthalmology and wound care.