In this study, we estimated cause-specific excess mortality in the four largest Nordic countries Denmark, Finland, Norway, and Sweden during the COVID-19 pandemic 2020–2022. During this period, all-cause mortality—as well as deaths attributable to COVID-19—was particularly large in Sweden in 2020, and in Denmark, Finland, and Norway in 2022. We here added several pieces of evidence which allowed us to shed light on the developments in the Nordics during this period. First, we observed a substantial under-mortality due to non-COVID-19 respiratory diseases and dementia that occurred across the four countries during the pandemic, not least for Sweden. Second, a substantial excess mortality due to cardiovascular disease was observed across countries and time, especially so for Finland and Norway, and for all countries towards the end of the period. Third, deaths due to cancer, diabetes, and external causes developed roughly in line with their pre-pandemic trends. A few exceptions to this pattern could be noted, including a steep increase in deaths due to diabetes in Norway and an under-mortality in deaths due to cancer in Sweden (the latter only in the models with ten-year trends).

Most major patterns that we have documented were qualitatively similar across the countries, but the differences in timing and magnitude were noticeable. In terms of deaths attributable to COVID-19, Sweden’s comparatively high mortality in 2020 may largely reflect this country’s relatively lenient restrictions. The 2022 surge in COVID-19 deaths in Denmark, Norway, and Finland might be largely related to the emergence of the more contagious Omicron variant, coupled with the easing of restrictions in the beginning of 2022. Sweden’s comparatively low mortality from COVID-19 in this year may reflect the strong pandemic impact in this country in 2020, which may have given rise to a mortality displacement over time, as well as a greater infection-induced immunity. Moreover, Sweden entered 2022 with a higher coverage of the booster vaccine dose than the other countries [34], further contributing to higher immunity. In total across the three pandemic years, Sweden experienced the largest number of COVID-19 deaths among the Nordic countries, even when considering these numbers in relation to population size. At the same time, Sweden had the smallest estimated all-cause excess mortality in relation to its population size when summing deaths over the three years. This comparatively small all-cause excess mortality reflects a larger under-mortality in dementia, a comparatively small excess mortality in cardiovascular deaths, and a possible under-mortality in cancer deaths in Sweden.

The under-mortality we observed in non-COVID-19 respiratory diseases for all the Nordic countries is in line with previous evidence from several countries including South Korea, the US, Australia, and several European countries [18, 21, 23,24,25, 27], and most likely reflects a reduced spread of infectious diseases, due to social distancing [35]. In addition, pathogenic competition may have played a role; becoming infected by COVID-19 might reduce the risk of contracting other respiratory infections, reducing the spread of non-COVID-19 respiratory diseases [36, 37]. The latter may be one reason why also Sweden, despite its less stringent restrictions, experienced substantial reductions in non-COVID-19 respiratory deaths.

The observed under-mortality in dementia is at odds with previous evidence from the US and Italy, which has instead suggested an excess mortality in dementia during the COVID-19 pandemic [17, 22, 23, 38]. Evidence from the UK suggested that there was an initial excess mortality in dementia, which was however followed by an under-mortality due to the same cause [24]. One likely explanation for the divergent results is that reporting of dementia deaths is subject to substantial measurement error [39]. For example, several autopsy studies have shown that deaths classified as being due to dementia often represent deaths due to respiratory disease such as pneumonia [40,41,42,43]. In line with this, a reduction in non-COVID-19 respiratory disease due to social distancing or other factors may have contributed to a reduction in deaths classified as dementia in the Nordic countries (while adherence to WHO guidelines for reporting of COVID-19 deaths [44], along with substantial testing for COVID-19, should have limited the likelihood of COVID-19 deaths being attributed to dementia).

Furthermore, those who died from COVID-19 were, at least in the early stages of the pandemic in both the Nordics and elsewhere, largely made up of older people residing in nursing homes [45,46,47]. This population has a very high prevalence of dementia [48, 49]. Consequently, the pandemic may also have caused a displacement from dementia deaths to COVID-19 deaths. While COVID-19 deaths peaked in Sweden in 2020, dementia deaths were primarily lowered in 2021 and 2022. This might reflect that many frail individuals with dementia died prematurely in 2020, hence reducing the population of individuals with dementia who would be at risk of dying in 2021 and 2022. Future work should assess how long-lasting this under-mortality in deaths classified as dementia will turn out to be, and whether similar patterns will be detected for Denmark, Norway, and Finland past their peak pandemic year in 2022.

The excess mortality due to cardiovascular disease which we observed for all countries is in line with evidence from several other contexts, including the US and several European countries [16, 17, 19, 20, 22,23,24, 27, 50,51,52,53,54,55,56], although evidence to the contrary exists as well [4, 16, 27, 55]. Several explanations for the observed patterns are possible. First, COVID-19 infection may itself be a risk factor for cardiovascular disease [57,58,59]. Moreover, healthcare visits for cardiovascular disease have in several countries been found to have decreased during the pandemic [12, 55, 60], a pattern that may represent a hesitancy to seek care, as well as a lower access to care [61, 62]. Worse prevention and management of cardiovascular disease due to disruption of healthcare services during the COVID-19 pandemic has also been documented [63]. More research will be needed to examine how such reductions in the quantity or quality of healthcare visits for cardiovascular disease may have contributed to increased cardiovascular mortality in the Nordic countries. Further examinations will also be necessary to assess the possible impact of other pandemic-related factors on cardiovascular mortality, such as social [64] and physical [65] inactivity, resulting from social distancing.

It is noticeable that excess mortality due to cardiovascular disease in Finland and Norway spiked already before these countries experienced any substantial numbers of COVID-19 deaths. Moreover, Sweden saw comparatively little excess mortality due to cardiovascular disease throughout the three years, and none in 2020, when the country was the most impacted by COVID-19. These observations suggest that factors other than COVID-19 infection, such as restrictions to access to care or hesitancy to seek care, may have been more important contributors to the excess cardiovascular mortality.

Several previous studies have found evidence of excess mortality due to diabetes during the COVID-19 pandemic [16, 17, 19, 22,23,24, 26, 66], although some countries were also found to exhibit under-mortality due to this cause [16]. Among the previous studies, one reported estimates for Denmark and Finland in 2020 [16]. Our estimates resemble these, but suggest that overall, during 2020–2022, the pandemic had little impact on mortality due to diabetes in Denmark, Finland, and Sweden. For Norway, however, an excess mortality due to diabetes was obtained – one that was much larger than what has been observed in most other contexts. (The finding might be an artifact of changes in the recording of deaths; see the discussion of limitations further down.)

Our results for cancer mortality, which showed no evidence of an impact, at least not in Denmark, Finland, and Norway, are in line with previous literature, which quite consistently has found no evidence of impacts of the COVID-19 pandemic on mortality due to cancer [16, 18, 19, 21,22,23,24,25]. Similarly for deaths due to external causes, our results confirm previous literature which generally suggests impacts to be small or nonexistent [18, 21,22,23,24] (a notable exception being deaths due to drug use [23]).

Our study has both strengths and limitations. As for the strengths, the Nordic register data are some of the most complete and accurate in the world [67], making our investigation an ideal case study for the complex mortality dynamics of the pandemic. Our approach accounts for time trends in mortality as well as differences in population age and sex structure over time and across countries—issues that, if unaccounted for, could substantially bias results in investigations of excess mortality [5, 6]. The addition of 2022 means that we covered the post-initial phases of the vaccination programs with the roll-out of booster doses, as well as the increased transmission of COVID-19 also among the vaccinated due to emergence of the Omicron variant.

Among the limitations of our study, the reporting of causes in death certificates comes with a degree of uncertainty, and practices for how deaths are assessed and reported differ across physicians and countries, as well as over time. Indeed, Fig. 1 showed substantial differences in cause-specific mortality before the pandemic, and some of these differences may reflect coding differences. Such coding differences may also have impacted the observed development during the pandemic. As one notable example, deaths classified as dementia are particularly prevalent in Finland, and it is possible that the development of excess mortality due to this cause is not fully comparable across Finland and the other countries.

Norway gradually moved to an electronic system for reporting causes of death during the pandemic years, an issue that might have influenced reporting in several ways, for example by nudging doctors to pick diagnoses that appeared in the predefined dropdown menu [28, 30]. In addition, this change involved an expansion of the maximum number of causes of deaths that were allowed to precede the immediate cause from two to three, meaning that causes that were previously considered too indirect to be included in the death certificate may now be reported as the underlying cause. This could potentially provide an explanation for the substantial increase in diabetes deaths recorded in Norway. However, pandemic impacts on healthcare utilization have also been forward as a possible explanation for the large number of recorded diabetes deaths in this country [30]. Future studies on excess mortality should try to shed more light on the role of coding differences across countries and over time, for example by utilizing data on multiple causes of death.

The rates of testing for COVID-19 varied across the Nordic countries, with Denmark conducting several times more tests per million people than the other countries [15]. This could potentially have led to a greater misclassification of COVID-19 deaths in the other three countries. It is important to note, however, that testing of hospitalized patients was strongly recommended across all countries [68].

Previous evidence has suggested that excess mortality estimates tend to be sensitive to the choice of reference period and other analytical choices [6, 69, 70]. Considering and comparing results from several models is therefore important. In this work, we focused on linear regression models, using a main model that assumed linear time trends over the ten-year period 2010–2019, and extrapolating the estimates to the pandemic years 2020–2022. As deaths are fundamentally recorded as counts, a count data model such as Poisson regression is in principle more adequate than linear regression. One disadvantage of count data models, however, is that there exists no gold standard for the calculation of prediction intervals [71]. A linear regression model offers a convenient way of calculating prediction intervals, and is useful in settings like ours, where populations are large, meaning that mortality rates can be seen as close to continuous. Our use of linear regression models is in line with a number of previous studies of excess mortality [70, 72,73,74,75,76,77,78], including our own investigations of all-cause excess mortality in the Nordics during the COVID-19 pandemic [5, 11]. The linear regression model flexibly accommodates non-linear trends, either by restriction of the study period to a shorter time frame (during which non-linearities may be negligible) or by transforming the outcome, such as by taking the logarithm of it, yielding a model with log-linear trends. Notably, the functional form of the linear regression model with log-linear trends resembles that of a Poisson regression model. As an additional check, we have applied Poisson regressions models (Supplementary Table S6). The results confirmed that the excess death estimates from Poisson regression were similar to those from linear regression with log-linear trends.

Compared to the results from our main linear regression model with ten-year trends, the model with five-year trends generally produced lower estimates of excess deaths for Denmark, Finland, and Norway, while larger for Sweden. In particular, applying the model with five-year trends resulted in a reduction of “Other” excess deaths in Denmark by around 1000–2000 per year, as compared to the main model, with prediction intervals being non-overlapping across the two models. The model based on log-linear trends (and Poisson regression) instead generally resulted in lower estimates of excess deaths for all countries than the main model. In particular for 2022, cardiovascular excess deaths in Finland, Norway, and Sweden decreased by about 1000 when using the log-linear model instead of the main model, with prediction intervals that were non-overlapping across this model and the main one. For estimates other than the aforementioned, prediction intervals were always overlapping when comparing estimates across the alternative models and the main one. However, substantial differences were still sometimes observed. Most notably, all-cause excess deaths differed by around 1000–2000 per year in Denmark and Sweden when comparing the model with five-year trends with the main model. Overall, the divergent results obtained from the different models underscore the importance of conducting sensitivity checks in studies of excess mortality, both with respect to functional form and the length of the reference period.