Melbourne: disease, prevalence and injury fatalaties data require updates

[carlhiggs]

Currently there are error warnings related to disease health transitions, prevalence and injury data when running SILO --- the latter two datasets in particular are missing for Melbourne, and the health transitions may require updating:

2025-08-11T15:13:11,611  INFO HealthTransitionTableReaderMEL:24 Reading health disease prob table from csv file
2025-08-11T15:13:11,761  WARN HealthTransitionTableReaderMEL:61 No enum constant de.tum.bgu.msm.health.disease.Diseases.all_cause_cancer
2025-08-11T15:13:11,762  INFO HealthTransitionTableReaderMEL:63 Finished reading health disease prob table from csv file.
2025-08-11T15:13:12,045  INFO PrevalenceDataReader:19 Reading prevalence data from ascii file
2025-08-11T15:13:12,048  INFO PrevalenceDataReader:55 Finished reading 0 prevalence data.
2025-08-11T15:13:12,052  INFO InjuryRRTableReader:28 Reading injury RR + fatalities data from ascii file
2025-08-11T15:13:12,055  INFO InjuryRRTableReader:72 Finished reading 0 prevalence data.
2025-08-11T15:13:12,046 FATAL PrevalenceDataReader:52 IO Exception caught reading prevalence data file: D:\projects\jibe\melbourne/input/health/base_prevalence_id_clean.csv
2025-08-11T15:13:12,047 FATAL PrevalenceDataReader:53 recCount = 0, recString = <>
2025-08-11T15:13:12,055 FATAL InjuryRRTableReader:69 IO Exception caught reading prevalence data file: D:\projects\jibe\melbourne/input/accident/injury_relativeRisks.csv
2025-08-11T15:13:12,055 FATAL InjuryRRTableReader:70 recCount = 0, recString = <>

Despite warnings, these errors are not Fatal; processing continues. But ultimately, we'll need to address this.

• remove all_cause_cancer (or otherwise address that warning)
• create prevalence dataset for Melbourne melbourne/input/health/base_prevalence_id_clean.csv
• create fatalities dataset for Melbourne melbourne/input/accident/injury_relativeRisks.csv

For now, I will use the Manchester prevalance dataset (dated 2 August 2025) as a placeholder. Manchester doesn't have the injury dataset in its gitlab location, but there is a copy in the JIBE working group folder at manchester/injury/injury_relativeRisks.csv. So that will be used for Melbourne as a placeholder.

[carlhiggs]

Regarding all_cause_cancer,
I did a regex for all_cause_cancer (^.*all_cause_cancer.*$) using notepad++ and used the mark/bookmark to note the lines, then delete them as per this StackOverflow advice. Noting that the file is under version control, and just in case, I took an offline back up too.

A new warning came up for kidney_cancer

2025-08-11T15:58:41,901  WARN HealthTransitionTableReaderMEL:61 No enum constant de.tum.bgu.msm.health.disease.Diseases.kidney_cancer

Rather than list every entry in health transitions that isn't found in the disease look up, it would be better to just make a list and report once at the end that these diseases were present in the data, but skipped as they weren't in the look up.

Regarding prevalence
Using a placeholder prevalence file, seemed to do the trick in temporary fix way:

2025-08-11T15:58:42,499  INFO PrevalenceDataReader:19 Reading prevalence data from ascii file
2025-08-11T15:58:43,698  INFO PrevalenceDataReader:55 Finished reading 646311 prevalence data.

regarding injuries/fatalities,
there were some issues with the placeholder--- the logger incorrectly refers to prevalence data (likely an artifact due to copied code from the prevalence reader; should update that), and the placeholder file didn't seem to be read despite apparently being located at the cited location:

2025-08-11T15:58:43,706  INFO InjuryRRTableReader:28 Reading injury RR + fatalities data from ascii file
2025-08-11T15:58:43,711  INFO InjuryRRTableReader:72 Finished reading 0 prevalence data.
2025-08-11T15:58:43,708 FATAL InjuryRRTableReader:69 IO Exception caught reading prevalence data file: D:\projects\jibe\melbourne/input/accident/injury_relativeRisks.csv
2025-08-11T15:58:43,711 FATAL InjuryRRTableReader:70 recCount = 0, recString = <>

so, I should debug that

[carlhiggs]

Re injury --- my mistake --- I located the accident folder within input/health/accident instead of input/accident; I see that's where its configured, and actually located for Manchester, including with tracking on gitlab. I'll move to using that as the placeholder.

I copied Manchester's accident folder for Melbourne, and added the following line in the SILO reference properties for Melbourne in the Health model section:

injury.rr.data.file = input/accident/age_gender_rr.csv

[carlhiggs]

After the above commit, and using the placeholder files (copies of Manchester's, for now), the errors have been resolved and log messages fixed:

2025-08-11T16:33:49,619  INFO HealthTransitionTableReaderMEL:24 Reading health disease prob table from csv file
2025-08-11T16:33:50,884  WARN HealthTransitionTableReaderMEL:71 Diseases not present in lookup: [malignant_melanoma, prostate_cancer, kidney_cancer]
2025-08-11T16:33:50,884  INFO HealthTransitionTableReaderMEL:73 Finished reading health disease prob table from csv file.
2025-08-11T16:33:51,085  INFO PrevalenceDataReader:19 Reading prevalence data from ascii file
2025-08-11T16:33:51,426  INFO PrevalenceDataReader:55 Finished reading 646311 prevalence data.
2025-08-11T16:33:51,428  INFO InjuryRRTableReader:28 Reading injury RR + fatalities data from ascii file
2025-08-11T16:33:51,437  INFO InjuryRRTableReader:72 Finished reading 48 injury RR + fatalities data.

The diseases list would include all_cause_cancer but I removed that manually from the local version of the file I am using; however, I'll revert that to the existing git version, so in future it will just skip that data. The improvement is that all diseases not present will be listed.

So, notwithstanding that Melbourne's data needs updating, this issue has been 'resolved' --- but because Melbourne's data ultimately does need updating, i'll leave it open until there are bespoke prevalance and injury data developed.

[carlhiggs]

After running the runHealthExposureOffline runner, using the Manchester-based placeholders for prevalance and injury, the following errors highlighted further updates to the health transition data that are required:

2025-08-12T19:25:43,214  WARN DiseaseModelMEL:29 Set up health model relative risk and disease state
2025-08-12T19:25:43,231  WARN DiseaseModelMEL:45 Health disease model end year:2018
2025-08-12T19:25:43,308  WARN DiseaseModelMEL:106 No health transition data for disease: parkinson
2025-08-12T19:25:43,309  WARN DiseaseModelMEL:106 No health transition data for disease: severely_injured_car
2025-08-12T19:25:43,309  WARN DiseaseModelMEL:106 No health transition data for disease: severely_injured_walk
2025-08-12T19:25:43,309  WARN DiseaseModelMEL:106 No health transition data for disease: severely_injured_bike
2025-08-12T19:25:43,310  WARN DiseaseModelMEL:106 No health transition data for disease: dead_car
2025-08-12T19:25:43,310  WARN DiseaseModelMEL:106 No health transition data for disease: dead_walk
2025-08-12T19:25:43,310  WARN DiseaseModelMEL:106 No health transition data for disease: dead_bike

So, the to do list now is:

• add health transitions for Parkinson's disease
• add health transitions for severely injured by each of car, walk and bike
• add health transitions for dead by each of car, walk and bike
• update Melbourne prevalence (melbourne/input/health/base_prevalence_id_clean.csv)
• update Melbourne fatalities ( input/accident/age_gender_rr.csv)

[carlhiggs]

Regarding Parkinson's disease, @StevePem advised in this thread that we are lacking data for this.

Looking into how this has been represented for Manchester, I believe this is not by LSOA per se but rather incident rates by age, sex and rural status. For example, for persons aged 71 across 561 LSOA areas there were only 4 distinct rates (by sex, and I suspect, by urban/rural classification of the LSOAs.

While we don't have Parkinson's incidence rate data for SA2s, we might be able to allocate this by age and sex, potentially stratified by rural status. However, one large-scale ecological population health study in Victoria (Ayton et al; https://doi.org/10.1016/j.parkreldis.2018.10.026) suggested that prevalence (not incidence, which wasn't studied) while associated with growing pulses (implicating pesticides, either produced naturally by the crops or used in their management) is not associated with rural areas per se. So if we can't locate data as a basis for urban/rural estimates, even just age and sex rates may be sufficient. I suspect this is something that would be achievable.

Parkinson's Australia links to commissioned reports by Mellick from 2024 including incidence and prevalence by sex for electoral divisions:
https://www.parkinsons.org.au/wp-content/uploads/2024/04/Ecosystem-of-Parkinsons-Part-3-Appendix-2.pdf

The data could be scraped from the PDF and assigned for SA2 regions based on approximate electoral divsion correspondence (some mis-match might be expected, that could be averaged over). For example, the electoral divsion of Cooper includes incidence estimates for 2021 by sex that look like the following:

The Division of Cooper (59 sq. km) consists of:
Darebin City Council,
part of the Moreland City Council, and
part of the Yarra City Council.
Inner Metropolitan - situated in capital cities and consisting of well-established
built-up suburbs.
In 2021 it is estimated that:
69 people developed Parkinson's (23 females and 46 males)
a total of 436 people (220 females and 216 males) were living with Parkinson's.

It might be that we could get age/sex incidence rates that we could adjust by sex using the geographic data relating to Greater Melbourne.

There are more statistics/data/estimates in the Mellick series of reports, e.g.
https://www.parkinsons.org.au/wp-content/uploads/2024/04/Ecosystem-of-Parkinsons-Part-3.pdf

In Table 23 they list incidence per 100k for Greater Melbourne at 0.00043 overall in 2020.

In part 2 of Mellick's report Table 2 summarises sex-stratified incidence by age group from Hirsch et al's 2016 meta-analysis. This incidence data was drawn upon by Mellick in their estimation of incidence by district. We could use this data in conjunction with the electoral division estimates data to derive age-specific incidence rate estimates.

Does this sound like a reasonable approach @BelenZapata85 ?

[carlhiggs]

Re Parkinson's, @BelenZapata85 recommends using GBD age and gender specific incidence and prevalance rate estimates, and adjusting these using the geographic incidence and prevalence.

• GBD 2018 estimates are here (a match for our 2018 target period)
• GBD 2021 estimates here (a match for the 2021 incidence and prevalence data by geography, if we use the data above from Mellick's report)

I believe though, the Mellick report just takes the age/sex demographics for each electroral district and applies the age/sex specific incidence and prevalance rates to get number estimates. In which case, these aren't spatially varying estimates and there is no need to use that data. For example, we could get SA2 age/sex demographics and apply this more directly, if that were relevant. But I suspect it isn't. As above, Manchester didn't have that degree of spatial variability in rates; I think variation may have been explained more by an urban/rural classification --- I will check.