Real-Time Biosurveillance Pilot in India and Sri Lanka
Nuwan WaidyanathaLIRNEasia
Email: [email protected]
http://www.lirneasia.net/profiles/nuwan-waidyanathaMobile: +8613888446352 (cn) +94773710394 (lk)
IEEE HealthCom'10Ambient assistive living for better health12th International Conference on eHealth, Networking, Applications, and Services
July 03, 2010Campus de la Doua, Lyon, France
This work was carried out with the aid of a grant from the International Development Research Centre, Canada.
www.lirneasia.net
Research Question: “Can software programs that analyze health statistics and mobile phone applications that send and receive the health information potentially be effective in the early detection and mitigation of disease outbreaks?”
Research overview
Specific Objectives
Evaluating the effectiveness of the m-Health RTBP for detecting and reporting outbreaks
Evaluating the benefits and efficiencies of communicating disease information
Contribution of community organization and gender participation
Develop a Toolkit for assessing m-Health RTBPs
www.lirneasia.net
Black arrows: current manual paper/postal system for health data collection and reporting
Red lines: RTBP mobile phone communication system for heath data collection and reporting
Sri Lanka Epidemiology and RTBP overlay
Reduce 15 - 30 day delays to Minutes
www.lirneasia.net
Black arrows: current manual paper/postal system for health data collection and reporting
Red lines: RTBP mobile phone communication system for heath data collection and reporting
India morbidity reporting and RTBP overlay
Reduce 7 - 30 day delays to Minutes
www.lirneasia.net
Data collection, Event detection, and Situational-Awareness/Alerting in RTBP
Skip the paper
www.lirneasia.net
Evaluation metric verticals and horizontals
Three verticals – data collection, event detection and reporting
Four layers – social, content, application, Transport
Arrows showing the Interoperability between layers and verticals
Objectively assess by calculating various indicators: costs, efficiencies, error rates, etc
Subjectively assess through interviews and simulations
Talk aboutthis
www.lirneasia.net
Our pilot in India and Sri Lanka
24 Health Sub Center Village Nurses
4 Public Health Center Sector Health Nurses, Health Inspectors, and Data Entry Operators
1 Integrated Disease Surveillance Program Unit of the Deputy Director of Health Services
Thirupathur Block, Sivagangai District, Tamil Nadu, India
12 District/Base Hospitals and Clinics
15 Sarvodaya Suwadana Center Assistants
4 Medical Officer of Health divisions & 1 Regional Epidemiology Unit
Kurunegala District, Wayamba Province, Sri Lanka
www.lirneasia.net
Data collection process
(1) Patient is received by the Nurse
(2) Nurse issues a diagnosis chit to patient fills in Name, Age, Gender, and OPD No
(3) Medical Officer fills in the chit with diagnosis and treatment
(4) Patient presents chit to pharmacy to receive medication
(5) Data Entry Operator digitizes and submits the data
1
2
3
4
5
www.lirneasia.net
mHealthSurvey Midlet by IIT-M
(a) Main menu
(b) Profile registration
(c) Retrieve locations
(d) Patient record screen I
(e) Patient record screen II
www.lirneasia.net
mHealthSurvy software design
J2ME: Built on Java 2 Micro edition
CDC: works with CDC 1.1 and above (JSR)
MIDP: works with MIDP 2.0 or above
GPRS: transport technology
Each record is 2kb and costs INR 0.01 or LKR 0.02 (USD 0.0002) i.e < USD 10 Handset/Month
Mobile phone around US$ 100
Tested on Nokia3110c, Motorola SLVR L7, Gionee v6900. Amoi A636, Sony Ericsson s302c
www.lirneasia.net
m-HealthSurvey Certification Exercise at the early stages
Sri Lanka India Benchmark0.00
10.00
20.00
30.00
40.00
50.00
60.00
70.00
80.00
90.00
100.00
Part IIIPart IIPart I
Country
Scor
es (
Min
=0;
Max
=10
0 )
Exercise Benchmark Sri Lanka IndiaPart I – installation and configuration (min) 12.00 10.75 17.48Part II – submit up to 6 records (min) 20.00 10.80 27.26Par III – standard operating procedures (points) 50.00 20.43 15.00
OutcomeCertified trainers ( > 90 points) 02 of 15 NilCertified Users (90 => points > 70) 13 of 15 04 of 23Uncertified (points <= 70) --- 19 of 23
Age 18 – 35No health trainingNo prior experience
Age 35 – 55Trained Health Workers10 – 25 yr experience
Average Country Scores
www.lirneasia.net
Y09-W
35
Y09-W
38
Y09-W
41
Y09-W
44
Y09-W
47
Y09-W
50
Y10-W
00
Y10-W
03
Y10-W
06
Y10-W
09
Y10-W
12
Y10-W
15
Y10-W
18
Y10-W
210
4000
8000
12000
Noisy Clean
Weeks
Cou
nts
Signal to Noise Ratios
Learning curve
FrictionAll MOH on board
New Year / Pongal
Unsupervised
Sin/Tam New Year
mHS v1.3
Y09-W
19
Y09-W
24
Y09-W
29
Y09-W
34
Y09-W
39
Y09-W
44
Y09-W
49
Y10-W
01
Y10-W
7
Y10-W
12
Y10-W
17
Y10-W
210
1000
2000
3000
4000
Noisy Clean
Week
Cou
nts
New Year / Pongal
HSC only
plus PHCConsequences
told
0.170.270.570.530300.58
0.18 0.40 0.31 0.04 0.07
www.lirneasia.net
Off-time vs Real-time
Y09-W
19
Y09-W
24
Y09-W
29
Y09-W
34
Y09-W
39
Y09-W
44
Y09-W
49
Y10-W
01
Y10-W
06
Y10-W
11
Y10-W
16
Y10-W
210
1000
2000
3000
4000
>= 1 day < 1 day
Weeks
Cou
nts
Y09-W
35
Y09-W
38
Y09-W
41
Y09-W
44
Y09-W
47
Y09-W
50
Y10-W
00
Y10-W
03
Y10-W
06
Y10-W
09
Y10-W
12
Y10-W
15
Y10-W
18
Y10-W
210
5000
10000
15000
>= 1 day < 1 day
Weeks
Cou
nts
No time routine work is top priority
Cannot complete before closing hours
Tech probs sharing phones
www.lirneasia.net
SNOMED-CT
Content ontology, semantics, syntax problemsProblem Examples
goal fever/jail fever/typus fever, dementia/memory loss, entric fever/typhoid fever, encephalitis/meningitis
body ache/body-ache, body pain,/body painmuscle weakness/weakness in musclestomach pain/pain in the stomach
Severe memory loss/memory loss
Nasal stuffiness or sneezingOver bleeding with abdominal pain
Not able to identify color
Oral pils, remove catheter, vaccination
Please specify details/specify symptoms
Diarrhoea/diarrhea, Vomiting/vommitting
BP 140/90, BP 120/100
muscle weakness/weakness in musclestomach pain/pain in the stomach
Fits/fits , cut/cuts
Tense Faint, fainted, fainting
Synonyms:
Insertion of symbols
Order of Words
Adjectives
Local language Leg vettuthal
Disjunctions
Long sentences
TreatmentInstructions
UK vs USA spellin
Test results
Multiple instances
Singular vs plural
LOINC
www.lirneasia.net
Fix the data collection shortcomings: noisy and off-time
0102
0304
0506
0708
0910
1112
13
0
3000
6000
9000
Noisy Clear
Week
Rec
ord
Cou
nts
01 02 03 04 05 06 07 08 09 10 11 12 130
2000
4000
6000
8000
Off-Time Real-Time
Week
Rec
ord
Co
unts
01 02 03 04 05 06 07 08 09 10 11 12 130
1000
2000
3000
Noisy Clear
week
Rec
ord
Cou
nts
01 02 03 04 05 06 07 08 09 10 11 12 130
1000
2000
3000
Off-Time Real-Time
Week
Rec
ord
Cou
nt
Need solutions
here
From: 01-Sep-2009 To: 30-May-2010
Case Records:
220000+
From: 01-Jun-2009 To: 30-May-2010
Case Records:
81000+
Noisy vs Cleandata
Real-Time vs Off-Timedata
www.lirneasia.net
Evaluation metric verticals and horizontals
Three verticals – data collection, event detection and reporting
Four layers – social, content, application, Transport
Arrows showing the Interoperability between layers and verticals
Objectively assess by calculating various indicators: costs, efficiencies, error rates, etc
Subjectively assess through interviews and simulations
Talk aboutthis
www.lirneasia.net
T-Cube Web Interface (TCWI) by Auton Lab
AD Tree data structure
Trained Bayesian Networks
Fast response to queries
Statistical estimations techniques
Data visualization over temporal and spatial dimensions
Automated alerts
www.lirneasia.net
Evaluation of TCWI
Replication study :: Sri Lankan Weekly Epidemiological Return (WER) reports published at www.epid.gov.lk notifiable disease counts tabulated by District was semi synthesized by distributing the weekly counts as daily counts taking day-of-week effect, gender distribution, and age representations.
Study the reliability and effectiveness :: significant events detected by T-Cube is compared with the ground truth and also weighed on the response actions or inaction
Competency exercise :: injected fake data over a period of 5 days and the subjects, unaware of the prefabricated events, were asked to detect most significant events
T-Cube Acceptance :: a questionnaire was designed based on the Technology Assessment Methodology (TAM) and was subject to TCWI users as well as health official associated with T-Cube who make decisions on whether or not to take action
Cost analysis :: compare the economic efficiencies and cost effectiveness between present detection/analyses system and T-Cube
www.lirneasia.net
Food poison spike as detected by spatial scan around Feb 15,2007 in Nuwara_Eliya, which was reported as outbreak by health department.
Spatial scan is run by 7 days windows size.
Replication study using Sri Lanka WER data 2007 - 2009
www.lirneasia.net
Another Food poison spike as detected by spatial scan around June 17,2009 in Nuwara_Eliya, the same location.
Spatial scan is run by 7 days windows size.
www.lirneasia.net
Dengue Fever Seasonal and spatial pattern
May 1,2007
Aug 30,2007
May 21,2008 April 15,2009 May 28,2009
www.lirneasia.net
First day an elevated global score noted, lead by region Kandy
4/14
Progression of Dengue Fever outbreak in April – June 2009
Spatial Scan global Score
4/15
Situation in Kandy intensified, together other regions
4/24
Southern Regions began to see increased cases
5/28
Southern region continue to see progression, while other region subsides
www.lirneasia.net
TCWI Competency Assessments with Synthetic data
Susceptible Exposed Infected Recovered
With a Network Flow
Injected 3 sets of data1) Notifiable disease :: Dysentery2) Other-Communicable disease :: ADD3) Syndrome :: Fever, Pain, RTI
Used “Epigrass” to generate synthetic data with a SEIR model
www.lirneasia.net
Evaluation metric verticals and horizontals
Three verticals – data collection, event detection and reporting
Four layers – social, content, application, Transport
Arrows showing the Interoperability between layers and verticals
Objectively assess by calculating various indicators: costs, efficiencies, error rates, etc
Subjectively assess through interviews and simulations
Talk aboutthis
www.lirneasia.net
Sahana Messaging/Alerting CAP/EDXL Broker by Respere
Single input multiple output engine; channeled through multiple technologies
Manage publisher /subscribers and SOP
Adopt PHIN Communication and Alerting Guidelines for EDXL/CAP
Relating the template editor with the SMS/Email Messaging module
Do direct and cascading alert from a regional jurisdictional prospective
Designing short, long, and voice text messages
Addressing in multi languages
www.lirneasia.net
Evaluation of Sahana Messaging/Alerting Module
Implementability of CAP :: ability to map the present policy and procedures to the Common Alerting Protocol (CAP) interoperable content standard for alerting and situational awareness messaging.
Comprehension exercise :: understand the ability of health workers to decipher CAP messages sent over SMS, Email, and Web
Usability :: determine the ability to create templates, create and issue alerts based on known scenarios
Cost analysis :: compare the economic efficiencies and cost effectiveness between present notification system and Sahana Messaging/Alerting Module
www.lirneasia.net
3 2 1 0Error0
50
100
Msg received via?
Affected locations?
Event?
Who issued?
Msg Identifier?
Msg priority?
Response actions?
Get more info?
Points
No
of s
ub
ject
s ga
ve a
nsw
er
Que
stio
ns
3 2 1 0Error0
50
100
Msg received via?
Affected locations?
Event?
Who issued?
Msg Identifier?
Msg priority?
Response actions?
Get more info?
Points
No
of s
ub
ject
s ga
ve a
nsw
er
Que
stio
ns
CAP SMS Alert/Situ-aware comprehension exercises
Outcomes
Everyone did quite well in the exercises except for 1 or 2 exceptional cases
Both India and Sri Lanka having trouble with msg-identifier; could be because msg-identifier getting truncated by the 160 char SMS constraint
Recommendation :: put msg- identifier in subject header (but may cutoff rest due to 160 char SMS); use the term “reference number” instead or both
Assessment design
Participants receive 4 SMS text with varying values of the CAP attributes
India = 23 and Sri Lanka = 19 health workers participated in the exercise
www.lirneasia.net
IN-Paper IN-mHealth LK-Paper LK-mHealth0.00
5,000.00
10,000.00
15,000.00
20,000.00
System delivery
System Admin/Support
Data center
Health facility
Health department
Health worker
Comparison of Functional Entity Costs in India and Sri Lankapresent paper based program vs mHealth program
System deliverySystem Admin/SupportData centerHealth facilityHealth departmentHealth worker
Country Program Categorizations
Cos
t (U
SD
) / D
istr
ict
/ Mon
th
Function
al Entit
y
Reforming the expenses to strengthen the disease control system
Introduce T-Cubeto shrink costs
Invest the shrinkcost in mHealthSurvey
for richer dataset
Strengthen thesitu-aware and
alerting with SAM
www.lirneasia.net
Conclusions mhealthSurvey is a worthy candidate for patient disease/syndrome digitization;
however, must be robust to minimize the noise and delays; need a better GUI if Medical Officers are to enter high volume real-time data opposed to a data entry clerk
Need a complete and comprehensive standardized disease syndrome ONTOLOGY perhaps a hybrid of SNOMED-CT and LOINC
T-Cube false alarm rates must be minimized through the iterative enhancement and machine learning
Sahana CAP Broker SMS, Email, and Web messaging has proven to be a winner for real-time adverse health event information dissemination but need Voice as well
Although value is seen in T-Cube Web Interface and CAP/EDXL Messaging The policies must be reformed to go beyond the century old paper based system to using ICT based event detection and alerting/situational-awareness
There should not be any institutional fears arising from the cost reductions instigated by the introduction of ICTs (e.g. RTBP) as is will still require the same work force
Before the cost benefits can take affect the laws and regulations must be changed to remove the paper and the storage cupboards that are government mandates