Draft of predictor section for the DHIS2 user guide


(Jim Grace) #1

Dear CBS community,

I've been rewriting the predictor section of the DHIS2 user guide. Attached
for your information and review is my current draft. I would be happy for
any suggestions for further improvement.

For those of you who received a draft yesterday on the old "Feed Back on
Predictor from the Surveillance Academy" thread, today's draft has some
improvements but is not very different.

Cheers,
Jim

predictors.pdf (446 KB)

···

--
Jim Grace
Core developer, DHIS 2
HISP US Inc.
http://www.dhis2.org <https://www.dhis2.org/>


(Jim Grace) #2

Hi All,

The new predictor documentation has been merged into the DHIS2 user guide,
with some small improvements since the draft I sent to this list. You can
see it at https://docs.dhis2.org/master/en/user/html/manage_predictor.html

Cheers,
Jim

···

On Fri, Sep 22, 2017 at 4:19 AM, jim@dhis2.org <jim@dhis2.org> wrote:

Dear CBS community,

I've been rewriting the predictor section of the DHIS2 user guide.
Attached for your information and review is my current draft. I would be
happy for any suggestions for further improvement.

For those of you who received a draft yesterday on the old "Feed Back on
Predictor from the Surveillance Academy" thread, today's draft has some
improvements but is not very different.

Cheers,
Jim

--
Jim Grace
Core developer, DHIS 2
HISP US Inc.
http://www.dhis2.org <https://www.dhis2.org/>

You are receiving this message because you are a member of the community DHIS2
Case-Based Surveillance Community
<https://knowledge-gateway.org/dhis2-cbs-community>.

View this contribution on the web site
<https://knowledge-gateway.org/_/fp8s3tsa>

A reply to this message will be sent to all members of DHIS2 Case-Based
Surveillance Community.

Reply to sender <jim@dhis2.org> | Unsubscribe
<leave.dhis2-cbs-community@knowledge-gateway.org>

--
Jim Grace
Core developer, DHIS 2
HISP US Inc.
http://www.dhis2.org <https://www.dhis2.org/>


(Jim Grace) #3

Hi Patrick,

Good questions!

• Do predictors rely on Tracker, Event Capture, or another source for data?

A predictor outputs an aggregate data value for a period. This value is
based on a formula containing any or all of the following types of data in
previous periods:
- Aggregate data, straight from the dataValue table. (Analytics need not be
run before making the prediction).
- Tracker data during a previous period
- Event capture data during a previous period

• Can Predictor be used to track both suspected and confirmed cases?

Yes. A predictor takes input from data elements, so it depends on how you
use the data elements. You can have data elements representing suspected
and/or confirmed cases, and use one or the other (or both) in the predictor
formula.

• Does Predictor form the foundation of the DHIS 2 outbreak detection and
notification system?

Yes. You can use a predictor to establish the limit of a value you consider
to be "normal" (non-outbreak), and then use a validation rule to test
whether new data is outside the normal range expected for that period.

• Does the Predictor system integrate with the mapping or case-based
surveillance functionalities of DHIS 2? If so, how does this work?

Yes, the predictor generates aggregate data values. After the predictor is
"run" to generate the values, analytics must be run before the value will
appear in mapping or pivot tables, etc. Once analytics is run, the
predicted value can be displayed using any analytics tools.

• Is there any plan to expand the functionality of predictors to
incorporate a predictive modelling component? Integration with case-based
surveillance would likely support calculation of transmission rates (from
date of disease/symptom onset, contact frequency and rate of contact
disease acquisition). This could allow charting of infectious disease
outbreaks and prediction/mapping of overall burden.

We are always interested in learning more about how we can improve DHIS2
features. I'm not sure how to translate what you say into requirements of
exactly what predictor calculations should do. How I can learn more about
what this would mean? (This is a question for anyone/everyone.)

Cheers,
Jim

···

On Thu, Nov 2, 2017 at 11:38 AM, Patrick Saunders-Hastings < psaunders-hastings@gevityinc.com> wrote:

Dear Jim,

Thank you very much for making this available. I was very interested to
read through this, and particularly appreciated the focus on event-based
surveillance as well as the flexibility of the Predictor system in setting
different thresholds across organizational units and in adjusting sampling
of past periods to reflect the health indicator and research question. I
had a few questions, listed below. Any insight you might be able to provide
on these would be greatly appreciated.

• Do predictors rely on Tracker, Event Capture, or another source for
data?
• Can Predictor be used to track both suspected and confirmed cases?
• Does Predictor form the foundation of the DHIS 2 outbreak detection and
notification system?
• Does the Predictor system integrate with the mapping or case-based
surveillance functionalities of DHIS 2? If so, how does this work?
• Is there any plan to expand the functionality of predictors to
incorporate a predictive modelling component? Integration with case-based
surveillance would likely support calculation of transmission rates (from
date of disease/symptom onset, contact frequency and rate of contact
disease acquisition). This could allow charting of infectious disease
outbreaks and prediction/mapping of overall burden.

Cheers,
Patrick

__________
You are receiving this message because you're a member of the community
DHIS2 Case-Based Surveillance Community.

View this contribution on the web site https://knowledge-gateway.org/
_/bfyfxbh4

A reply to this message will be sent to all members of DHIS2 Case-Based
Surveillance Community.
To reply to sender, send a message to psaunders-hastings@gevityinc.com.
To unsubscribe, send an email to leave.dhis2-cbs-community@
knowledge-gateway.org

--
Jim Grace
Core developer, DHIS 2
HISP US Inc.
http://www.dhis2.org <https://www.dhis2.org/>


(Jim Grace) #4

Hi Patrick,

Thanks, that sounds quite interesting!

I will put this back to the community: Is there work underway or upcoming
that would benefit from this type of predictive modeling? (See the
preceding email from Patrick Saunders-Hastings on this thread.)

Patrick, the reason I ask the community for more feedback is to help set
our implementation priorities. As I said, we are always interested in
learning more about how we can improve DHIS2. But also, as you might
imagine, we generally have more feature requests than time to implement
them. So we tend to be field-driven in setting our priorities. Do you know
of any specific organizations or projects that might want to use this
analysis should it be in DHIS2? To your knowledge is this used by any
countries, or CDC, or NGOs?

Another approach would be to do predictive modeling outside of DHIS2 at
first, but using DHIS2 data. The data could be exported, predictive
modelling applied, and the results imported back into DHIS2. If this were
done for example in a pilot project, we might learn from the results what
kind of integration into DHIS2 would be most beneficial.

Meanwhile, if you have readings to suggest, please post them to this list.
I would like to learn at least some more about this and maybe others would
too. (Maybe start a new email thread with a subject such as "Predictive
Modeling in DHIS2". Feel free to repeat any or all of the information you
just sent.) Thank you!

Cheers,
Jim

···

On Tue, Nov 7, 2017 at 10:25 AM, Patrick Saunders-Hastings < psaunders-hastings@gevityinc.com> wrote:

Hi Jim,

Thank you very much for your quick and very helpful response. With regard
to my question about predictive modelling, this arose from one of my own
research interests in emerging disease preparedness and response planning.

My doctoral research involved the construction and implementation of a
mathematical model to chart pandemic influenza transmission against
hospital-resource capacity in Canada. However, one of the big points of
frustration in infectious disease modelling is the knowledge that
assumptions used to inform model inputs will likely not reflect the actual
parameters of the next disease outbreak. This is especially true in
lower-resource environments where such empirical research is more scarce.
This leads to uncertainty in disease estimates and response planning, and
there is currently an unmet need for predictive models that are able to
adapt to data as it becomes available in real time. This is where I see a
potential space for DHIS 2.

Put simply, the case data currently being tracked in DHIS 2 could be used
to develop increasingly accurate estimates of key disease parameters. Dates
of estimated transmission, development of symptoms and alleviation of
symptoms could together calculate latent period (time between exposure and
infection), incubation period (time between exposure and onset of clinical
symptoms), and infectious period (duration of infectiousness – could be
approximated as duration of symptoms). Meanwhile, expanded capability to
track the number of contacts with whom a given case interacts — and the
proportion of these that become infected — would allow estimation of
disease transmissibility and transmission rate.

In terms of what this would mean for predictor calculations, this would
represent a departure from relying on past data counts to detect abnormal
increases and potential outbreaks. Instead, it would allow next-generation
calculations of how incidence and prevalence of a given disease may
increase or decrease over coming days, weeks and months (as well as the
implications of these changes for resource demand). Generator calculations
could be built using estimates of disease attack rate, contact rate,
transmissibility and duration of infection (from case-based data). And most
exciting from my point of view: automated calculation functionality would
support these estimates becoming increasingly accurate and reflective of
actual disease parameters as more cases were recorded. These developments
would also represent an important step towards meeting an unmet need for
predictive modelling in lower-resource environments.

I hope that this has proved helpful, and would be happy to chat or suggest
some background reading if you would be interested in discussing further.

Cheers,
Patrick

__________
You are receiving this message because you're a member of the community
DHIS2 Case-Based Surveillance Community.

View this contribution on the web site https://knowledge-gateway.org/
_/xnpm09mb

A reply to this message will be sent to all members of DHIS2 Case-Based
Surveillance Community.
To reply to sender, send a message to psaunders-hastings@gevityinc.com.
To unsubscribe, send an email to leave.dhis2-cbs-community@
knowledge-gateway.org

--
Jim Grace
Core developer, DHIS 2
HISP US Inc.
http://www.dhis2.org <https://www.dhis2.org/>


(Edem Kossi) #5

Hi all
This is an interesting discussion and I think Patrick is pointing to
possibilities to say for instance a given ongoing outbreak will infect x
number of people in n days if nothing is done. This is quite interesting
because telling decision makers the outbreak would claim this number of
deaths in x days if a response is not quickly in place speaks more than
just says saying there is this number of cases or deaths.
Having said that, given that we have so many priorities we need to
prioritise them. But I do believe even if it is not possible to implement
this in dhis2 now the community can learn how to do these predictive
analysis using different tools. In so doing we can gradually learn and see
how to better implement it in DHIS2.
Best

Edem


(Scott Russpatrick) #6

Hi Patrick et all,

Latent, incubation, and infectious periods should be calculable via program
indicators and the aggregation set to average if you are doing case-based
data capture and the tracker program is configured properly. The next step
would be tieing those program indicators to predictor data elements to
produce the forcasted values, which is possible in the latest build of
2.28. I think all the functionality currently exists for this.

This would essentially enable the predicted values to continuously update
based upon new patients while still factoring in historical data. Happy to
chat through this or take a look at those readings.

Best,

Scott

···

On Wed, Nov 8, 2017 at 10:47 PM, ekossi@gmail.com <ekossi@gmail.com> wrote:

Hi all
This is an interesting discussion and I think Patrick is pointing to
possibilities to say for instance a given ongoing outbreak will infect x
number of people in n days if nothing is done. This is quite interesting
because telling decision makers the outbreak would claim this number of
deaths in x days if a response is not quickly in place speaks more than
just says saying there is this number of cases or deaths.
Having said that, given that we have so many priorities we need to
prioritise them. But I do believe even if it is not possible to implement
this in dhis2 now the community can learn how to do these predictive
analysis using different tools. In so doing we can gradually learn and see
how to better implement it in DHIS2.
Best

Edem

On 8 Nov 2017 20:36, "jim@dhis2.org" <jim@dhis2.org> wrote:

Hi Patrick,

Thanks, that sounds quite interesting!

I will put this back to the community: Is there work underway or upcoming
that would benefit from this type of predictive modeling? (See the
preceding email from Patrick Saunders-Hastings on this thread.)

Patrick, the reason I ask the community for more feedback is to help set
our implementation priorities. As I said, we are always interested in
learning more about how we can improve DHIS2. But also, as you might
imagine, we generally have more feature requests than time to implement
them. So we tend to be field-driven in setting our priorities. Do you know
of any specific organizations or projects that might want to use this
analysis should it be in DHIS2? To your knowledge is this used by any
countries, or CDC, or NGOs?

Another approach would be to do predictive modeling outside of DHIS2 at
first, but using DHIS2 data. The data could be exported, predictive
modelling applied, and the results imported back into DHIS2. If this were
done for example in a pilot project, we might learn from the results what
kind of integration into DHIS2 would be most beneficial.

Meanwhile, if you have readings to suggest, please post them to this list.
I would like to learn at least some more about this and maybe others would
too. (Maybe start a new email thread with a subject such as "Predictive
Modeling in DHIS2". Feel free to repeat any or all of the information you
just sent.) Thank you!

Cheers,
Jim

On Tue, Nov 7, 2017 at 10:25 AM, Patrick Saunders-Hastings < > psaunders-hastings@gevityinc.com> wrote:

Hi Jim,

Thank you very much for your quick and very helpful response. With regard
to my question about predictive modelling, this arose from one of my own
research interests in emerging disease preparedness and response planning.

My doctoral research involved the construction and implementation of a
mathematical model to chart pandemic influenza transmission against
hospital-resource capacity in Canada. However, one of the big points of
frustration in infectious disease modelling is the knowledge that
assumptions used to inform model inputs will likely not reflect the actual
parameters of the next disease outbreak. This is especially true in
lower-resource environments where such empirical research is more scarce.
This leads to uncertainty in disease estimates and response planning, and
there is currently an unmet need for predictive models that are able to
adapt to data as it becomes available in real time. This is where I see a
potential space for DHIS 2.

Put simply, the case data currently being tracked in DHIS 2 could be used
to develop increasingly accurate estimates of key disease parameters. Dates
of estimated transmission, development of symptoms and alleviation of
symptoms could together calculate latent period (time between exposure and
infection), incubation period (time between exposure and onset of clinical
symptoms), and infectious period (duration of infectiousness – could be
approximated as duration of symptoms). Meanwhile, expanded capability to
track the number of contacts with whom a given case interacts — and the
proportion of these that become infected — would allow estimation of
disease transmissibility and transmission rate.

In terms of what this would mean for predictor calculations, this would
represent a departure from relying on past data counts to detect abnormal
increases and potential outbreaks. Instead, it would allow next-generation
calculations of how incidence and prevalence of a given disease may
increase or decrease over coming days, weeks and months (as well as the
implications of these changes for resource demand). Generator calculations
could be built using estimates of disease attack rate, contact rate,
transmissibility and duration of infection (from case-based data). And most
exciting from my point of view: automated calculation functionality would
support these estimates becoming increasingly accurate and reflective of
actual disease parameters as more cases were recorded. These developments
would also represent an important step towards meeting an unmet need for
predictive modelling in lower-resource environments.

I hope that this has proved helpful, and would be happy to chat or
suggest some background reading if you would be interested in discussing
further.

Cheers,
Patrick

__________
You are receiving this message because you're a member of the community
DHIS2 Case-Based Surveillance Community.

View this contribution on the web site https://knowledge-gateway.org/
_/xnpm09mb

A reply to this message will be sent to all members of DHIS2 Case-Based
Surveillance Community.
To reply to sender, send a message to psaunders-hastings@gevityinc.com.
To unsubscribe, send an email to leave.dhis2-cbs-community@know
ledge-gateway.org

--
Jim Grace
Core developer, DHIS 2
HISP US Inc.
http://www.dhis2.org <https://www.dhis2.org/>

You are receiving this message because you are a member of the community DHIS2
Case-Based Surveillance Community
<https://knowledge-gateway.org/dhis2-cbs-community>.

View this contribution on the web site
<https://knowledge-gateway.org/_/btbbzy2n>

A reply to this message will be sent to all members of DHIS2 Case-Based
Surveillance Community.

Reply to sender <jim@dhis2.org> | Unsubscribe
<leave.dhis2-cbs-community@knowledge-gateway.org>

You are receiving this message because you are a member of the community DHIS2
Case-Based Surveillance Community
<https://knowledge-gateway.org/dhis2-cbs-community>.

View this contribution on the web site
<https://knowledge-gateway.org/_/h2tf79j7>

A reply to this message will be sent to all members of DHIS2 Case-Based
Surveillance Community.

Reply to sender <ekossi@gmail.com> | Unsubscribe
<leave.dhis2-cbs-community@knowledge-gateway.org>

--

*Scott Russpatrick*

DHIS 2 Implementation Adviser

Health Information Systems Programme (HISP)

Department of Informatics

University of Oslo

Phone: +47 93 03 51 84 | Skype: scott.russpatrick

E-mail: s <karolitl@ifi.uio.no>cott@dhis2.org | scottmr@ifi.uio.no