Manitoba COVID report has changed
jeanpaulrsoucy opened this issue · 11 comments
Report now being released as PDF here: https://www.gov.mb.ca/health/publichealth/surveillance/influenza/index.html
Will be updated Friday mornings, so have to add a new entry in data_sources.csv
with the appropriate info.
The changes to the Manitoba report will affect data quality.
First, the report has removed cumulative numbers, or least the cumulative numbers being reported do not represent the entire time period. Part of the report suggests the numbers are cumulative since July 3, 2022. However, the "total deaths" are listed at 159, even though changes to cumulative deaths have been 209 (2256 - 2047) up to November 5 (the end date of the previous report). The sum of weekly deaths (an underestimate of total deaths) over the same period was 90, so it's not using that number either. We could use the difference between this number and whatever the next report gives for the number of new deaths next week, but are lost this week (or if the if the window of reporting they use changes).
For deaths, which were at the province level anyway, we will have to use PHAC epidemiology update data going forward. (Assuming it continues to be updated)
Cases are more difficult. In previous reports, the cumulative at week t was not quite cumulative week t-1 + weekly cases at week t, but they weren't so far off either. For future weeks, we could try to use the change in the cumulative numbers reported, assuming the reference date doesn't change, but this is a big if. We could also use the weekly numbers and use the difference in the cumulative value later reported in the PHAC update as "unknown" health region cases, but of course, this number could end up being negative. We could also give up on the health region level altogether and just report at the provincial level using the PHAC data.
I think for cases, we can use the weekly values for this week and evaluate in future weeks whether cumulative differences are feasible.
Next week, we can check if using the new partial cumulative numbers is viable for cases and deaths. There are fewer reasons to give it try with deaths, since the reporting delay is longer and the PHAC numbers are reliable and come out on the Monday following the release of the report on Friday.
Note that during this update, I also identified an issue with the previous death data for MB. For the previous report, I was using the weekly reported deaths rather than the cumulative death counts, which led to an undercount of deaths over time since deaths occurring in earlier reporting periods were not counted in the weekly death counts. This has been fixed.
Another report came out this week, some observations:
- The "anchor point" for cumulative numbers remains the same (July 3, 2022).
- According to the data notes: "Cumulative data includes updates from start of the current season to end of the reporting week; due to reporting delays or amendments, the sum of weekly report totals may not add up to cumulative totals."
- The cumulative cases number at the top of the report didn't get updated, so this should be ignored. The sum of the cumulative cases by health region is 5944.
- The weekly cases at the top of the report is 314, which is equal to the sum of the weekly cases reported by health region.
- The difference between the cumulative value from this week and last week is 312, which is NOT the same as the sum of the weekly values (314). However, it is not that different either.
For cases, I think I will use the weekly case values for the first week (i.e., whenever the seasonal anchor point changes) and then use differences in the cumulative values thereafter. If I change my mind later, I can always use the weekly values.
For deaths, I will continue to use the PHAC values, since there is no real advantage to using the Manitoba data except a three-day time advantage.
Major issue with MB cases introduced by Friday's report fixed by ccodwg/Covid19CanadaDataProcess@c264d54 in 1ee2028.
Need to update utils/report_mb.R
to facilitate updates using the new data structure.
There was an error in the dates for the last MB report. This should be fixed in the next update.
Interestingly, the cumulative total for the province (6480) is not the same as the summed cumulative total for the health regions this week (6478). But weekly case numbers match.
Besides the discrepancy noted above, it appears the new update script worked perfectly with this week's update.
One thing to verify is that if health region incidence gets really low, we're going to start seeing "<5" in the cases_weekly column for health regions. I should update the script to accept strings for the health regions.
The update process for MB has been basically figured out for now. Can always re-open this issue or open a new one if there are further developments.
It is good that we are using cumulative values for case reporting, since Manitoba did not bother producing a report to cover epi week 51, which was missed over the holiday period. The current report covers only week 52, meaning the cases in week 51 would be missed. It remains to be seen whether the death time series in the PHAC data will contain both weeks of data individually, or just a 2 week catch-up.