Automatic case classification for existing SORMAS diseases [13]

[StefanSzczesny]

The classification of a case should automatically be done based on defined rules. These rules in-clude a combination of specific symptoms, epidemiological data, vaccination/immunization, lab test results and more.

The definitions are provided in the SORMAS data dictionary and will be implemented with a rule system (probably not a rule engine) that combines different criteria and can be extended to be customizable by admin users later. https://docs.google.com/spreadsheets/d/1BA_RjM-ZhxFrzpAJdv0UVGhicHXNP6JnPBA-xzdrQKs/edit#gid=466434037

The classification is done whenever any of the above data is changed. It is possible to manually re-classify cases.

The relevant diseases are: EVD, Lassa fever, HPAI, CSM, Measles, Cholera, Yellow fever, Dengue Fever, Monkeypox, and Plague.

[MartinWahnschaffe]

@hzi-braunschweig

From my understanding the automatic case classification will need the following additional fields:

• "disease suspected by clinician" - a checkbox in the symptoms form
• "no test possible" a checkbox next to the samples list of the case in the case form

I'm not sure about the following points:

1. What is the exact defintion of "epidemiological link" in terms of the data we are collecting in SORMAS? Is it a contact to a confirmed case? Or also a contact to a probable/suspected case?
2. What does "suspected from a verified rumor" exactly mean ("probable" for EVD)
3. What is the exact defintion for epidemiological risk factor (also in comparison to point 1).
4. Is the difference between "a person with fever" (HPAI) and "a person with sudden onset of fever" (CSM)
5. How is "detection of fourfold increase in yellow fever IgM, or IgG antibody titres between acute and convalescent serum samples, or both" represented in SORMAS? Will this need two separate samples that are compared? This would mean that we also have to enter the exact test result (yellow fever IgM).
6. What about positive postmortem liver histopathology? Is "liver hisopathology" a new test type we need?
7. What about the case classification rules for monkeypox and the different plague types?

[MartinWahnschaffe]

Result from yesterdays meeting at HZI.

I have also updated the case classifications in the data dictionary: https://docs.google.com/spreadsheets/d/1BA_RjM-ZhxFrzpAJdv0UVGhicHXNP6JnPBA-xzdrQKs/edit#gid=466434037

Case Data

• User should be able to do a manual classification
• When a manual classification is overriden by an automatic classification (due to changed data) the responsible user should be informed (e.g. with a task to check the classification)
• keep "log" of classifications with automatic and manual comments (e.g. what rules were applied)
• add information-icon tooltip to classification to list the classification rules for the current disease
• Add "Dengue type": Dengue Fever, Dengue Haemorrhagic Fever, Denuge Shock Syndrome (Dengue)
• Rename HPAI to "Influenca (new subtype)" (short: "New flu")

Symptoms

• Rename "Rash" (lesions) to "Vesiculopustular rash" (Monkeypox)
• Rename "Skin rash" to "Maculopapular rash" (e.g. Dengue)
• Add "Meningeal signs" (CSM)

Epi Data

• Epi link (all diseases): "Did the case have contact with a confirmed case during the incubation period?". The user should be able to select a source case. This would basically select/create a contact the current person had with the source case. Automatism for this: System should automatically assign the contact, but not check the checkbox
• "Exposure to disease infected neighborhood within incubation period" (Yellow, Cholera, Plague)
• "Laboratory processing of blood or body fluids from a confirmed case while not wearing appropriate PPE or without using standard biosafety precautions" (EVD)
• "Percutaneous (e.g., needle stick) or mucous membrane exposure to blood from a case" (EVD)
• "Direct contact with a dead body while not wearing appropriate PPE" (EVD)

Samples

• (Add field to case (in sample area): Checkbox "No (additional) sample can be taken" (e.g. EVD))
• Add info "To add tests to this sample it has to be marked as received first" to the bottom of the sample edit view
• Add "tissue" to sample material (e.g yellow fever)

Sample Tests

• Replace Rapid test with: "Antigen detection test (Rapid test)"
• Rename Serum antibody titer to "IgM serum antibody"
• Add specific test types "Yellow fever IgM serum antibody", "Dengue fever IgM serum antibody" and "West nile fever IgM serum antibody"
• Add specific test types "Yellow fever neutralizing antibodies", "Dengue fever neutralizing antibodies" and "West nile fever neutralizing antibodies"
• Remove test type ELISA (replace existing with IgM)
• Join test types to "PCR / RT-PCR"
• Add test type "Histopathology"
• Add test type: "Yersinia pestis antigen test"
• Add test type "IgG serum antibody"
• Add checkbox "Four-fold increase in antibody titer" for IgM and IgG tests. Only editable when there are at least two samples (add tooltip).
• make sure test types are sorted alphabetically

Questions

1. Adding the "No (additional) sample can be taken" to the samples section in the app would be a lot of work and changes to the existing design. On the other side adding this checkbox to another section (e.g. case data) would not make a lot of sense. I'm wondering whether we really need this checkbox. An alternative would be to simply use "Deceased suspected case having an epi-link to a confirmed case" as the rule for "probable".
2. For CSM I the classification had no definition for the duration of the incubation period (used for epi link in probable cases). I have used the 10 days that are defined in SORMAS, okay?
3. We don't have a symptom "Bloody diarrhea". Instead there is a symptom for "Diarrhea" and a symptom for "Blood in stool". I'd interpret the combination of both symptoms as bloody diarrhea for EVD suspected cases, okay?
4. For dengue we currently don't have "probable". I was wondering whether the combination of a suspected case (not necessarily deceased) with "exposure to disease infected neighborhood within incubation period" would make sense?
5. We don't have "severe dehydration" (Cholera suspected). Instead only "dehydration" - used for EVD, Lassa and Cholera. Should we (A) rename this field, (B) use it for the classification as is or (C) add a second field?
6. For Lassa suspected the rule contains "contact with excreta or urine of rodents". In the epi data section we already have "Contact with: Rodents or their excreta". I expect we can use this?
7. "Absence of yellow fever immunization within 30 days before onset": We currently have a field for vaccination and vaccination date. How does this translate? Must the vaccination take place before the 30 days before onset?
8. Finally the influence epi link part. This one is tricky:

This is the original set of epi links for HPAI:

a. Close contact (within 1 meter) with a person who is a suspected, probable, or confirmed HPAI case; b. Exposure (e.g. handling, slaughtering, de-feathering, butchering, preparation for consumption) to poultry or wild birds or their remains or to environments contaminated by their faeces in an area where H5N1 infections in animals or humans have been suspected or confirmed in the last month c. Consumption of raw or undercooked poultry products in an area where H5N1 infections in animals or humans have been suspected or confirmed in the last month d. Close contact with a confirmed H5N1 infected animal other than poultry or wild birds e. Handling samples (animal or human) suspected of containing H5N1 virus in a laboratory or other setting

Result from yesterday and my thoughts: a. Same room with a probable or confirmed case b. Exposure (e.g. handling, slaughtering, de-feathering, butchering, preparation for consumption) to animals (or their remains) likely to have been infected by the disease c. Consumption: Not sure how to change this one. d. and e. This would be no longer necessary (see b.), right?

Besides this question, this is what we are currently asking for in SORMAS (see screenshot attached). I'm wondering whether we need all of this / what should be used/changed for the epi links above.

"Please indicate an answer regarding ALL animals (live or dead) the person had direct exposure to during the incubation period."

• Eating raw or undercooked poultry
• Exposure to poultry or domesticated birds
• Expose to sick/unexplained dead poultry/other domest. birds
• other animals
• exposure to wild birds

[MartinWahnschaffe]

Answers

1. OK. As long as this rule is only applicable to deceased suspect cases, it may be acceptable to simplify as suggest.
2. OK
3. OK
4. Please rephrase to “exposure to neighborhood where confirmed cases occurred within incubation period”
5. Simply use “dehydration” throughout , without the adjective “severe”
6. OK
7. OK
8. Use the following rules:
   • Close contact (within 1 meter) with a probable or confirmed case
   • Exposure (e.g. handling, butchering) to animals (or their remains) in an area where infected animals have been confirmed in the past month
   • Consumption of raw or undercooked animal products in an area where infected animals have been confirmed in the past month
   • Handling samples (animal or human) suspected of containing the virus without adequate personal protection

In addition:

• rename "eating raw or undercooked poultry" to "Eating raw or undercooked meat or other animal products"
• add a details field
• remove "exposure to poultry or domesticated birds"
• rename "exposure to sick/unxeplainedly dead poultry/other domest. birds" to "... animals"
• remove "other animals"
• remove "exposure to wild birds"

[MartinWahnschaffe]

Maté and have discussed my idea of a case classification log. A problem with this would be that it could sometimes become quite big and result in unnecessary data this needs to be transfered to the mobile app.

From our point of view the essential data for a mobile user is:

• who changed the case classification
• when did this happen
• optionally a comment field could be added
• the comment field should be prefilled with "Changed from ..." when the clasification is changed
• comment field should also explain the rules applied (when done automatically) ".. because ..."

If it becomes necessary to look into the full history of classification changes this could be implemented in the web app at some point based on the data from the history tables.

[MartinWahnschaffe]

• [x] Research: Check existing open source rule engines; find a clean, maintainable and adaptable way to define the classification
• [x] build model to define classification
• [x] apply automatic case classification whenever a relevant field value changes (not whenever a case is saved)
• [x] OR: add "lastAutomaticClassification" to the entity level to compare against. Not to the api, because we don't want to force an app update right now
• [x] build classification description from model
• [x] saving hospitalization and epi data should call saveCase (see symptoms). We probably need to remove saveHospitalization and saveEpiData methods in the facades.
• [x] show classification changed info in UI (when saving) -> works, but not for hospitalization and epi data, see above
• [x] I18n of classification description
• [x] add property to enable automatic classification for server (for now disable by default)
• [x] automatic classification also needs to be done when saving sample tests -> call onCaseChange in saveSampleTest and also when deleting sample tests or samples
• [x] make OneOfCriteria child class of XOfCriteria
• [x] rename all classes to Classification...Criteria
• [x] enable four fold increase for other test types, change sample test criteria to always include the desired test type and change classification logic + tests (IGM & IGG)
• [x] ~find a way to do generic unit tests for this (e.g. parameterize all relevant values: https://www.tutorialspoint.com/junit/junit_parameterized_test.htm)~
• [x] ~add classification rule information to mobile app~ see #783
• [x] add classification date and user fields to case form in app
• [x] fix bug that sometimes prevents classification rules popup from opening
• [x] ~think about a way to move classes back to the backend~ -> keep in API
• [x] add some comments (e.g. to compact criteria)

For comment:

• [x] add clickable info button next to case classification that opens a popup explaining the rules for the automatic classification of the respective disease (use existing description)
• [x] comment should not be added to UI for now
• [x] classifying user should be replaced by another label field stating "System" when the case has been automatically classified

Testing:

• [x] add two test methods per disease, one that tests the desired behaviour and one that rules out false positives
• [x] false positives should be ruled out by e.g. setting all symptoms to yes apart from the ones that trigger automatic case classification
• [ ] Cholera test not working right now because of EJB injection issue for tests -> Person has to be a parameter of the eval method, remove backend call from criteria

Add classifications for

• [x] EVD
• [x] Lassa fever
• [x] Yellow fever
• [x] Dengue Fever
• [x] Influenca (new subtype)
• [x] CSM
• [x] Measles
• [x] Cholera
• [x] Monkey pox
• [x] Plague

[MartinWahnschaffe]

Three possible rule engines:

1. Drools (https://www.drools.org/): Uses a script language for the rule definitions. Comes with a whole set of tools and also a process engine. Here is a short introduction article: https://medium.com/@ryanjollyyoung/why-should-i-use-drools-ba80be3b5311
2. OpenRules uses Excel for rule files. I couldn't find the source code though and the documentation looks very old-schoolish.
3. "Decision Model and Notation" (DMN): A standard of the OMG (https://en.wikipedia.org/wiki/Decision_Model_and_Notation, https://docs.camunda.org/manual/7.9/reference/dmn11/decision-table/). Uses decision tables that are stored in an XML format. Flowable (java business process engine: https://www.flowable.org/) comes with a rule engine for this.

At the current point I wouldn't suggest to add a rule engine, though. It would simply add a lot of overhead to the whole thing. The way to go is to put the logic for the automatic case classification in its own class and give it a clean interface. If needed this can be easily replaced at a later point in time.

The most important part will be unit tests that cover all possible rule paths.

[MateStrysewske]

@hzi-braunschweig Is "Malaise" equivalent to "Fatigue/Weakness"? -> Yes New Influenca: Which test type is meant with "neutralization antibody test"? -> New test type needed, see #780

[MartinWahnschaffe]

Adjust case classification visual presentation:

• [x] add an bold "AND" at the end of each line (see picture)
• [x] put a background behind all lines of a rule to make clear that they belong to e.g. "probable" (see picture. Grey backgrounds should probably have some transparency).
• [x] rename unexplained bleeding ... to "Bleeding or bruising"

Fix bugs (tested on symeda server):

• [x] ~Automatic classification not fully working. Example: EVD suspect worked. But afterwards removing symptom or fulfilling criteria for probable or confirmed did not work.~
• [x] ~date of classification is not shown~

[MateStrysewske]

I can't reproduce the bugs on the symeda server.