Introduction

The Central Statistics Office (CSO) has a policy of careful and considered investigation of a data source before employing it to inform official statistics. In Ireland, people have three months to register a death, meaning it can be many months after the event before official statistics on deaths can be produced. This does not include any deaths that are referred to the Coroner’s Court. However, given the need for more timely data as a result of the COVID-19 pandemic, the CSO began to explore experimental ways of obtaining up-to-date mortality data.

Due to the Irish custom of holding funerals within two to three days following death, online death notices are usually placed in a fast and efficient manner, providing a valuable crowd-sourced means of tracking deaths. The notices are placed close to 'real time', the average length of time between date of death and publication is about 1.1 days. When compared with the statutory time limit of three months¹ for the registrations of deaths in the State, this is a timelier data source for monitoring trends in mortality.

The CSO published the first analysis of death notices using the website RIP.ie in July 2020. The initial analysis explored the use of death notices posted on RIP.ie to measure trends in mortality, finding a strong correlation between death notices and official CSO mortality statistics. Subsequent analysis in November 2020 and April 2021 continued on this work, expanding the excess mortality analysis as well as updating the data to include the most recent RIP.ie death notices.

Analysis of death notices was conducted as far back as 01 October 2019, to include the last month before the first global cases of COVID-19 were notified. This updated release covers the period October 2019 to December 2021 providing a full 27-month analysis of trends in mortality using death notices.

Measuring Excess Mortality

The COVID-19 pandemic has highlighted the challenges in obtaining timely statistics on mortality and in particular, excess mortality (deaths over and above what would be expected under normal circumstances). Measuring the impact of such events on mortality trends can play an important role in policy and response efforts and in determining any additional impact on mortality related to the COVID-19 pandemic and its effects.

However, measurement of excess mortality is not straightforward and attributing trends to specific factors, in this study COVID-19, is particularly challenging. Measuring excess mortality, specifically where COVID-19 was a factor, would necessitate cleaning, classifying and linking each RIP.ie notice with the corresponding General Register Office (GRO) record which contains information about causes of death or contributory factors. This would need to be done for several previous years to identify and adjust external factors that may influence the number of notices placed. Adjustments might be required for many reasons including differing levels of expatriate usage of RIP.ie, broadband penetration, potential changes in road traffic fatalities due to changes in traffic volumes and/or traffic patterns and a reduction in influenza deaths due to fewer personal interactions.  

What is Excess Mortality and how is it measured in this report? Excess mortality measures the number of deaths over and above what would be expected under normal circumstances. Expected deaths are measured by calculating the average number of deaths for the same period in previous years. To get excess mortality we compare the deaths (or in this case death notices) in one period with the average of previous periods. In that sense excess mortality is an estimate.  In this release we look at a range of expected deaths (in the four years prior to the pandemic, 2016 to 2019) based on death notices published on RIP.ie. In some studies calculations of excess mortality look at an average of a greater number of years (e.g., a five year or 10-year average). The CSO has recently been providing Eurostat with estimated weekly deaths based on the methodology used in this release. Aligning this release with the work Eurostat has done on excess mortality allows comparisons with other European countries. Annual deaths are increasing over the last decade due in part at least to an ageing population. It is important to differentiate between this natural demographic effect and the effect of the pandemic. This is another reason to look at more recent years to establish expected mortality. It should also be noted that the calculation of mortality, including excess mortality, is based on deaths that occur in the Republic of Ireland. RIP.ie includes some deaths that occur in Northern Ireland and therefore these are removed. What can estimates of excess mortality tell us? Estimates of excess mortality look at all causes of mortality during a period of observation, so the estimates of excess deaths are different to measurement of the number of deaths due to a particular cause (e.g., COVID-19 confirmed deaths). While COVID-19 confirmed deaths increased the level of excess mortality in the period, other causes may have had lower or higher impacts than in previous periods and none of these impacts can be disaggregated. Death notices, while having the advantage of timeliness, lack information on cause of death, and the age of the deceased, which is necessary to accurately profile mortality patterns over time. COVID-19 confirmed deaths include people who died with and of COVID-19, which results in a different period total to all cause excess mortality, though the peak months are the same in both measures. Another difference between the two measures – i.e., excess deaths and deaths involving COVID-19 – is excess deaths can be negative, where deaths due to COVID-19 cannot be. For example, when compared with the average for the three years prior to the pandemic, there were seven months over 2020 and 2021 where there were less death notices than average. These months were January, February, June, July, November and December in 2020 and March 2021.

 

Get the data: PxStat VSD01 (monthly to 2018), PxStat VSA09 (annual to 2020)

Figure 2 examines monthly mortality figures by calendar year, with observed RIP.ie data forming the basis for cases from January 2020 to December 2021 and the remaining data coming from CSO mortality figures (discussed below). While there are observable seasonal peaks, there is significant variance year on year. A clear increase can be seen in the level of death notices observed in April of 2020 which contrasts with observed dips in mortality levels in April of all previous years. A total of 3,528 death notices were recorded in April 2020, which is over 900 more than the next highest number of deaths recorded in any April in the previous five years (2,615 in April 2018). The number of death notices decreased in the summer months of 2020 before beginning to rise again from September to December 2020, which is in line with the trends seen in previous years. Death notices in the first two months of 2021 show an increase compared with the same months in previous years. 

In January 2021 3,965 death notices were recorded which is 460 more than the next highest year, 3,505 in 2018. Death notices in the last six months of 2021 also show an increase compared with the same months in previous years however the increase is significantly less than the increase seen around both April 2020 and January 2021.

It is worth noting there was no evidence of influenza viruses circulating in the community in Ireland during weeks 13 and 14 2021 (weeks ending 04/04/2021 and 11/04/2021) or during weeks 40 2020 to week 12 2021 (weeks ending 04/10/2020 to 28/03/2021). In week 12 2021, (ending 28/03/2021), the Health Protection Surveillance Centre (HPSC) reported that "no influenza, Respiratory syncytial virus (RSV) or Acute respiratory infection (ARI - excluding COVID-19) outbreaks were notified to HPSC during the 2020/2021 season to date".

In week 1 2022, the most recent report available at time of publication, states that "Influenza activity remained low in Ireland during week 1 2022 (ending 09/01/2022)" and also "During the 2021/2022 season (weeks 40 2021 - 1 2022) one death in a notified influenza A(H3) case was reported to HPSC during week 45 2021." The most recent reports on the 2021/2022 Influenza Season from the Health Protection Surveillance Centre (HPSC) can be accessed here. 

Figure 3 plots observed and expected death notices over a 12 month period.

One method for calculating expected deaths is to take the average number of deaths over a number of years prior to the pandemic and compare the period of interest to that average. In this instance we calculate the average of the last four years before the pandemic (2016 to 2019) and compare this average to the data in 2020 and 2021.

The expected numbers of death notices are then subtracted from the observed numbers to arrive at the excess mortality. (Figure 4)

Analysis of death notices for the period January 2020 - December 2021 (noting that 11 March 2020 was the date of the first reported death due to COVID-19 in Ireland) provides an estimate for excess mortality of 3,533. This compares to the number of confirmed COVID-19 deaths over the same period of 5,912. The difference in these two figures highlights that although there is a strong relationship, COVID-19 deaths are not by default excess deaths. Both figures are important as it allows analysis of both COVID-19 deaths and the overall impact of the pandemic through excess deaths. 

To demonstrate the difference in choosing a different time period and the impact that this has on excess mortality estimation, further analysis was carried out to estimate the excess mortality for the 12-month period March 2020 - February 2021. This period includes the two most significant peaks of confirmed COVID-19 deaths in Ireland. Excess mortality in this reference period was 2,128 and the number of confirmed COVID-19 deaths over the same period was 4,294.

Death Notices and Gender

The gender of the deceased was inferred from death notices using forenames, personal pronouns, possessive pronouns and relative terms such as ‘beloved mother of’ and ‘husband to’. For the period under investigation, 52% of death notices were for males while 48% were for females. Deaths reported by the CSO had a gender split of 51.1% male and 48.9% female for the years 2007 to 2017 inclusive. Of the 5,952 confirmed COVID-19 deaths⁴ up to 04 January 2022, 53.5% were male, while 46.5% were female.

Methodological Notes

To calculate excess mortality, we subtract the number of death notices we would have expected to see in the absence of the pandemic from the number of death notices we observed. 

Observed Death Notices

Each death notice pertaining to a date of death between 01 October 2019 and 02 January 2022 (the period under investigation) was read, cleaned and classified. Cleaning involves first classifying deaths as within Ireland or abroad and then removing:

• Death notices pertaining to expatriate deaths
• Death notices pertaining to Northern Irish citizens
• Nulls – these are empty records which contain no information, a result of the web-scraping process
• Duplicates, which are so rare as to be negligible

Classification Process

For the first three releases in this series, classification of location of death was undertaken by a team who read every individual notice and manually classified the death notice as referring to a death in Republic of Ireland, Northern Ireland or elsewhere abroad. For this release a new method of classification was developed by a CSO statistician using a machine learning classification model. The models were trained on the results of the manual classification work done during the previous releases. These models when applied to any newly scraped data will provide an appropriate classification for each new death notice. For consistency the classification model has been applied to all RIP data contained in this release. The result of this change means that some historical figures will have been updated. Furthermore, there is no limit on when a notice can be uploaded to RIP and as such new notices may have been added recently that relate to deaths that occurred months or very occasionally, years ago. In all cases a date of death is specified and additional notices are only included if they relate to the reference period of this release.

 

Expected Death Notices

Expected death notices are calculated as the death notices we would be have expected to see during the period of interest in the absence of the pandemic. Previous releases calculated expected death notices in two different ways:

• The average of the last two years
• The average of the last three years

For this release we have moved to a single average over the four years prior to the pandemic (2016 - 2019). The reason for this change was to align the baseline average with the methodology used by Eurostat in calculating excess deaths. By aligning this release with the Eurostat methodology, it is possible to compare excess deaths with other European countries. The Eurostat data can be accessed here.

 

Excess Mortality and the Choice of Period of Interest

Our analysis covers the period beginning in January 2020, to account for the impact of COVID-19 as the first recorded death was 11 March 2020. We examined the periods March to June 2020 inclusive in our first release, March to September 2020 inclusive in our second and March 2020 to February 2021 in the third release.

For our period of interest in this release, we cover the months January 2020 to December 2021, which includes two peaks associated with COVID-19. While it is preferable to avoid beginning a reference period around a peak of winter mortality, where seasonality is observably established, this release is intended to provide analysis based on full calendar years. Due to the novelty of this illness, it is still not clear if there is a seasonal pattern of deaths associated with it or where it might peak. 

1. In addition, now that 2021 has ended, we can show an estimate of deaths for 2021 for comparison with earlier calendar years, with the caveat described above. It should be further noted that the rise in deaths from one year to another is not automatically considered to be excess deaths. Indeed, the number of deaths annually, per calendar year, has followed an increasing trend since around 2010. Between 2010 and 2018, the average annual rise in the number of deaths was 397. The largest fall was 252, between 2013 and 2014. The largest rise was 875, between 2014 and 2015.

2. We continue the series by showing single value for excess mortality for the period beginning in January 2020 and ending in December 2021. These excess figures are based on the four year pre-COVID19 average, the years 2016 to 2019. The first confirmed case of COVID19 in Ireland was in late February 2020.

Comparison of Death Notices to Deaths by Occurrence

Deaths by Occurrence measure deaths by the date on which they occurred. Death notices were also analysed by date of death. At the time of writing, the CSO had published monthly deaths by occurrence up to and including December 2018. Furthermore, the CSO had published annual provisional deaths by occurrence for 2019. A monthly series of 24 months was constructed from the provisional annual data for 2018 and 2019, extending the monthly series of deaths from October 2014 to September 2019. The annual provisional figures were apportioned to months using death registrations with the General Register Office (by date of death) for the years 2018 and 2019. This was on the basis that this was the best indication available at the time of writing of a monthly distribution for the years 2018 and 2019. It is important to have an accurate monthly distribution as we are calculating an excess death notice figure for a period of months – in this case, the twenty-four-month period January 2020 to December 2021.

Although the CSO deaths by occurrence are not being used to calculate excess mortality at this time, they are singular as an official benchmark to measure death notices against. As noted in the HIQA report, there is an undercount of death notices relative to deaths by date of occurrence the further one goes back in time. Reasons for this phenomenon may be explained by market penetration, regional variations, use of competing services, broadband usage and IT literacy. Other complexities ignored in this analysis are the different timing of the influenza seasons in different countries and the mortality of expatriates, which could impact the proportion of expatriate death notices in different months. It is noted that the resulting expected death notices are lower than if we were to adjust for this undercount. Part of the discrepancy in 2018 and 2019 may be attributable to the fact that the annual data from which the months are derived, are provisional.

The ad hoc and experimental nature of this updated analysis is noted; however, death notices continue to competently track registrations, albeit with a mean lag of one day compared with a mean lag of 63 days for registrations.

For further COVID-19 related information go to the CSO COVID-19 Information Hub

---

Footnotes

¹Registering a Death in Ireland

²The analysis of COVID-19 confirmed deaths by date does not include numbers for probable COVID-19 deaths (which are included in totals provided by the National Public Health Emergency Team). It was not possible to provide analysis of probable COVID-19 deaths by date. For full analysis of COVID-19 confirmed and probable deaths please see the COVID-19 Deaths and Case Statistics releases.

³Cases where the place of death was unclear or unstated have been combined with cases where the circumstances of death are potentially identifiable. These low-frequency instances include: 1) there were a small number of deaths at residential facilities for retirees of religious vocations that are not regulated by HIQA; 2) deaths due to accident; 3) deaths at residential facilities caring for people living with mental health issue.

⁴HSE - Weekly report on COVID-19 deaths reported in Ireland prepared by HPSC

⁵Register of Older Person's Facilities (CSV download)

---