Identifying signature features of epidemic diseases from 19th century all-cause mortality data

Abstract

Deadly epidemics leave distinct marks on all-cause mortality. When cause-specific health data are unavailable, studies of all-cause mortality may be necessary for understanding epidemic and pandemic diseases in history. Here, we identify and catalog every major epidemic in Denmark during the 100-year period between 1815 and 1915, based on a recently digitized and compiled data set of all 4 million burials during the period. Although the data set lacks specific information on cause of death, we were able to determine plausible etiology for the majority of 418 identified mortality crises that had more than 50 excess deaths. Epidemiologic methods, data analysis, consultation of historical sources, and investigation of the signature features of age patterns, seasonality, timing, and geography were used. The identified epidemics included, among others, pandemic influenza, cholera outbreaks in 1853 and 1857, and annually repeating epidemics during the period 1826-1832. Although these epidemics have been discussed elsewhere, our work presents a different view of these epidemics, based solely on all-cause mortality. Some of the identified epidemics were caused by pathogens that still affect us in modern times. In low-income modern settings for which representative population health data may be unavailable, the use of mortality data to determine the signature features may guide policy and improve future mitigation strategies.

signature features, excess mortality, all-cause mortality, historical, Denmark, mortality, crises, pandemic flu, cholera

Introduction

Pandemic and epidemic diseases have afflicted humanity since time immemorial. Studying the dynamics of historical epidemics is important because it may provide insight into developments in modern-day infectious disease dynamics. To guide modern-day policy and pandemic response, it is necessary to understand the signature features of disease, such as age patterns, geography, and seasonality.¹ Although modern pandemics and epidemics are rich sources of data on such phenomena, a comprehensive examination requires analysis of historical examples, as well.

The study of epidemic diseases in history has long been hampered by the lack of detailed time-series data on morbidity or mortality, with high spatial resolution. However, in recent years, access to historical health data has undergone a significant transformation: large-scale digitization projects of public records have created extensive historical health data sets and vastly accelerated the discovery of unexplored population data.^2,‑5 These have enabled studies of 19th century epidemic diseases with a level of detail that was not previously possible. Historical pandemic and epidemic diseases, such as cholera, smallpox, dysentery, plague, and influenza, have been studied extensively^6,‑9; however, such studies are often limited to urban populations for which cause-of-death data are available (eg, well-known sources like the London Bills of Mortality).¹⁰ Hence, patterns of morbidity and mortality in rural populations remain poorly understood. Furthermore, the epidemiology of childhood infections, such as measles, pertussis, and scarlatina (or scarlet fever), in the 19th century have received only little attention in the literature, despite being common causes of death at the time.¹¹

A major source of historical demographic information is parish registers.^12,‑14 A review of the quantitative methods for exploring such data can be found in the work of Hinde.¹⁵ Recently, a large collection of Danish parish registers, with a nearly complete national coverage between 1814 and 1920, was digitized and transcribed for the Danish National Archives by the genealogical corporation Ancestry. These records include births, marriages, and burials, with information about parish, name, age, and sex. This data set allows for a detailed epidemiologic study over a 100-year period and can provide important insight into infectious diseases in 19th century populations.

In this study, we use this newly transcribed data set of parish register data to identify and categorize the most important infectious disease mortality crises in 19th century Denmark. Our purpose is to present the data, our methodology for documenting these mortality crises, and how we define the signature features of these events. We highlight several deadly epidemics, including some that have received only little attention in the literature. By a combination of excess mortality calculations from time series analysis, data clustering, and consultation of historical literature, we identify, distinguish, and compare all identified mortality crises and their signature features across geography and time.

Methods

A unique source of 19th century mortality data

We accessed digital transcribed copies of 100 years of burial records from 19th century parish registers, stored by the Danish National Archives.¹⁶ These registers contained information on all births, deaths, baptisms, marriages, and confirmations within all Danish parishes. From 1814, parish registers in Denmark were kept in 2 copies (by the pastor and a parish clerk), which were later compared and reconciled.¹³ Transcription of the reconciled documents was done by the genealogy corporation Ancestry, yielding a digital data set of all individual entries in the Danish parish registers. Burial records were available for the period 1814 until 1920, with some geographic differences in the earliest and latest years available in data. Because of these differences, our analysis focused on the period from 1815 to 1915.

Scans of the original source material are available online, courtesy of the Danish National Archive.¹⁶ We used these to verify data integrity and to obtain additional information not present in the transcribed data.

Data processing

In-depth details on data processing and analysis can be found in the Supplementary Material online. Here, we briefly describe our approach. We extracted date of death for 4 million individuals from burial records across almost 2000 parishes in Denmark. For 94% of burials, the records included information on the person’s age. In the case of stillbirth, the age field of the parish register typically contained this information as text. However, initial investigations of the scanned source material revealed that some stillbirths were registered with an age of 0 years in the original source material, with only a margin note mentioning the stillbirth. This made a consistent distinction between stillborn and children aged 0 years impossible. Therefore, we consider stillborn and 0-year-old infants as a single age group for our analysis. Furthermore, we grouped burials in the following age groups: 1-14, 15-39, 40-60, and ≥60 years. Entries with missing age were included in analysis of total mortality but not in age-specific analysis.

Geographic organization

We grouped parishes into counties according to reference information on the historical administrative geography of Denmark¹⁷ by linking parish names to geographic reference data and grouping based on county bounds. In the referenced geographic information, the city of Copenhagen was separate from the surrounding Copenhagen County and, for purposes of our analysis, the city was considered separate from the county.

Major changes in the geographic organization of counties and their constituting parishes occurred in the period investigated, and we accounted for this by grouping data in separate groups before and after a given change (Supplementary Material). Time series of daily burials for all counties in 5-year age intervals are available online at https://github.com/PandemiXcenter/SignatureFeatures19thCentury.

Mortality baseline and identification of mortality crises

We constructed time series of daily deaths for each county. To calculate excess mortality, we computed the baseline of expected mortality for each day as the mean of the daily number of deaths on the same date in the preceding and succeeding 12 years, yielding a 24-year mean of the surrounding years of data. Baselines were calculated for each county, both for the entire population and for the specified age groups. Excess mortality was calculated as the difference between the baseline and observed number of deaths. To determine the z score (ie, standard score), the SD of the baseline data was also calculated.

Because major mortality events influence the baseline mortality of surrounding years substantially, we used an iterative process in which all data points with a z score above a factor of 3 were removed and a new baseline and z score were calculated from the remaining data. This process was repeated until no data points exceeded the threshold of 3 z scores above the baseline.

After establishing the baseline this way, we defined mortality crises (ie, periods of substantial excess mortality) as periods with at least 1 day with a z score > 3.^15,18 Startpoints and endpoints for each period were determined as the date when the number of deaths first dropped below 2 z scores above the baseline for more than 7 days.

Population data

To calculate per-capita excess mortality, we compiled a table of population counts at the county level for the investigated period, based on census data¹⁹ from 14 censuses between 1801 and 1916, approximately once per decade. Data from 2 of these are presented in Table 1. Exponential interpolation was used to estimate annual population counts in years between censuses. Population counts of specific counties in southern Denmark were not estimated, because there was a lack of data due to the cession of the duchies of Schleswig-Holstein from Denmark in 1864.

Table 1

Key statistics of Danish counties.^a

County	Geographic region	Population in 1834	Population in 1901	No. of mortality crises	Total excess deaths
Copenhagen (city)	Zealand	119 292	400 575	53	11 090
Copenhagen (county)	Zealand	62 365	172 962	28	3823
Frederiksborg	Zealand	67 054	90 476	19	1763
Holbæk	Zealand	63 868	98 325	21	4305
Sorø	Zealand	57 184	94 422	22	4511
Præstø	Zealand	70 199	103 293	25	4532
Maribo	Lolland and Falster	66 186	105 021	41	6796
Bornholm	Bornholm	24 645	40 889	7	886
Odense	Funen	90 155	151 544	33	4068
Svendborg	Funen	76 907	128 241	17	1755
Hjørring	Northern Jutland	61 503	119 385	15	1375
Thisted	Northern Jutland	43 171	71 438	11	1448
Aalborg	Northern Jutland	59 948	128 656	13	1901
Viborg	Western Jutland	54 449	106 608	9	1131
Ringkøbing	Western Jutland	54 013	111 474	19	1757
Ribe	Western Jutland	52 698	95 682	11	741
Randers	Eastern Jutland	63 709	118 586	9	1039
Skanderborg	Eastern Jutland	39 738	b	9	607
Aarhus	Eastern Jutland	76 670b	186 440	23	2513
Vejle	Eastern Jutland	59 781	125 523	16	1624
Sønderborg	Southern Jutlandc			3	948
Haderslev	Southern Jutlandc			9	1083
Åbenrå	Southern Jutlandc			2	285
Nordborg	Southern Jutlandc			2	140
Tønder	Southern Jutlandc			1	58

^a Data from additional censuses are omitted from this table for brevity; however, the selected census data show the population increase over the studied period, as well as the urbanization illustrated by Copenhagen city increasing more than 3-fold, whereas rural counties (eg, Maribo County) increased less than 2-fold.

^b Before 1824 and after 1867, Skanderborg County was part of Aarhus County. The 1834 population for Aarhus County is the sum of the population for Aarhus and Skanderborg counties in the 1834 census.

^c Conceded in 1864. No census data from these regions were used in this work.

Identifying signature features of mortality crises

For each identified mortality crisis, we determined specific characterizing the following “signature features”: total excess mortality, age-specific excess mortality, duration, seasonality, date with most deaths, and sex distribution as a fraction of male burials out of all burials in the period. From visual inspection of the mortality time series, we focused our analysis on mortality crises with ≥50 excess burials. This allowed us to identify mortality crises of local significance, such as the Baltic Sea flood of 1872, for which our method identifies 60 excess deaths in Maribo County.

To analyze age patterns, we calculated the fraction of total excess mortality arising from each age group and used Gaussian mixture modeling (GM) to group the age patterns into 6 distinct clusters (clusters A-F), each representing a distinct age distribution (see the Supplementary Material for details).

Identifying the cause of mortality

To determine likely causes of each mortality crisis, we checked the scanned parish registers for mentioned causes of deaths. Because there was no requirement for the pastors to document the cause of death, such information was only rarely available, and typically only in the cities. Therefore, we also consulted annual medical reports by the Danish Royal Board of Health,²⁰ in which all Danish physicians and surgeons were required, from 1803, to document epidemic diseases in their local area. Up to 1862, these reports were purely qualitative; they later also included tabulated numbers of infectious disease cases and deaths. The medical reports were ordered by period and geography, allowing us to easily identify relevant reports. Furthermore, we searched scanned Danish newspapers maintained at the website of the Royal Library²¹ for mentions of disease outbreaks or disasters during periods of interest. We looked for descriptions of epidemics in the region where and time period when excess mortality occurred, using search words (in Danish) such as “syge,” “epidemi” (meaning ill, epidemic), as well as more specific words such as “skarlagen,” “blodgang,” and “feber” (scarlet, dysentery, fever).

Results

Cataloging all mortality crises of 19th century Denmark

In the 100-year period from 1815 to 1915, we identified 418 mortality crises with ≥ 50 excess deaths at the county level. Details of each crisis are given in Table S1, along with the identified signature features of each crisis, such as total excess mortality, age-specific mortality, and more. Note that each of the 418 mortality crises refers to a crisis in a particular county. Hence, crises that occur simultaneously across county borders are represented by multiple entries in the table.

Figure 1 illustrates the results of the GM clustering of the age structure. A complete figure of the results is provided in the Supplementary Material (see Figure S5 in that file). In Table 1, the geographic distribution of mortality crises by county is presented numerically, and Figure 2 shows a graphical illustration of all mortality crises. The 418 mortality crises accounted for a total of 60 179 excess deaths over the 100-year study period, constituting ~ 1.5% of all the 4 million burials recorded in Denmark in the data set. Graphical overviews of each mortality crisis were produced to aid our analysis, with the 20 deadliest shown in the Supplementary Material.

Figure 1

The results of clustering using Gaussian mixture modeling. A) Fractions of all excess mortality due to the age groups of 0-year-old and stillborn infants on the horizontal axis, and the fraction due to children aged 1-14 years on the vertical axis. Each crisis is shown as a dot, with color corresponding to the cluster to which the age pattern of the crises belonged. The mean value of each cluster is shown as open circles. B) Excess mortality due to age groups 40-59 and ≥60 years is shown. Note that each crisis is shown in both panels. The 15-39 years age group is omitted from this figure but can be seen in the Supplementary Material along with additional age-group pairings. C) The mean value of each cluster is shown, illustrating the proportion of excess mortality due to age-group-specific mortality in each cluster.

Figure 2

Overview of all identified mortality crises. Each of the mortality crises identified is indicated as a horizontal line with rounded ends from the start to the end of the crisis. The coloring is based on the GM cluster to which the mortality crisis belongs, according to the age pattern. The size is scaled based on the estimated number of excess deaths, as indicated in the figure key.

Based on the identified signature features, the deadliest mortality crises were easily identified as epidemics and pandemics that are documented in scientific literature and by contemporary physicians. These include a series of extremely deadly rural epidemics during the harvest seasons of 1826-1832, the cholera epidemics in 1853 and 1857, scarlet fever epidemics in 1858, and pandemic influenza in 1889-1892 and 1900. A complete tally of excess deaths from these epidemics across all counties is given in Table 2. In the following sections, these epidemics are presented in detail along with our rationale for grouping these mortality crises together.

Table 2

Overview of groups of major mortality crises identified in 19th century Denmark with the same plausible cause.

Geographya	Time period (months, year)b	Total excess mortalityc	Age group (years)d	GM cluster (A-F)	Results subsection titlee
Maribo, Præstø, Sorø, Holbæk, Copenhagen, Frederiksborg, Svendborg, Odense, Sønderborg  (primarily eastern Denmark)	August-October 1827-1831, and spring 1832	10 974	Adults (≥15)	Mostly E	The 1826-1832 Epidemics
Maribo, Præstø, Sorø, Holbæk,  Frederiksborg, Svendborg, Odense, Haderslev  (primarily eastern Denmark)	Spring 1829	2720	Children (1-15)	C	The 1826-1832 Epidemics
Copenhagen, Copenhagen City, Aalborg, Aarhus, Hjørring,  Frederiksborg (cities)	July-October 1853	5154	Adults (≥15)	D	The Cholera Epidemics
Sorø (province towns)	September-November 1857	431	Adults (≥15)	D	The Cholera Epidemics
Thisted, Aalborg, Viborg, Randers, Aarhus, Skanderborg  (Jutland)	October 1857 to May 1858	2241	Children (1-15)	Mostly B	Scarlet Fever
All counties	December 1891 to February 1892	6462	Elderly (≥60)	F	Pandemic Influenza
All counties except Bornholm,  Copenhagen, Randers, Ribe,  Ringkøbing, Viborg, Aarhus	March-May 1900	1874	Elderly (≥60)	F	Pandemic Influenza

^a Counties in which the identified mortality crises occurred and general description.

^b The period with the highest excess mortality.

^c The total of the excess mortality (ie, number of excess deaths) of the constituting mortality crises.

^d Based on GM clustering as well as a visual inspection of the age-specific mortality of the crises.

^e The title of the subsection of the main text in which each group is described in detail.

The 1826-1832 epidemics

Denmark experienced a period of high mortality from 1826 to 1832, particularly on the islands of Zealand, Lolland, and Falster. The mortality was particularly elevated during the months of August, September, and October.

We identified several mortality crises with this seasonality and timing, many of which share a similar age distribution corresponding to high mortality among adults, grouped as GM cluster E. The last of these crises began in 1831 but were found to continue into 1832, with excess mortality in Maribo County lasting until August 1832. These mortality crises occurring during 1826-1832 totaled 10 974 excess deaths, almost 200% more than the baseline mortality for the same period. Figure 3 shows excess mortality of the afflicted counties.

Figure 3

The 1826-1832 epidemics. Excess mortality for the counties of Holbæk, Præstø, Sorø, and Maribo. The inset displays the number of burials for these counties, split by burials of individuals below and above 15 years of age. Vertical bars representing the months of august, September, and October are highlighted in gray to illustrate the seasonality of the harvest epidemics.

The signature features of these epidemics allowed us to distinguish them from another major epidemic in the spring of 1829. The crises in the spring of 1829 primarily affected children aged 1-15 years, clustered as GM cluster C. This difference in age pattern and seasonality is illustrated in the inset in Figure 3. An investigation of contemporary medical reports revealed that these deadly spring mortality crises occurred during outbreaks of measles and scarlet fever, whereas the rural harvest epidemics were of unknown etiology and likely augmented by a food security crisis.²²

The cholera epidemics

Unlike other European countries, Denmark experienced only 1 nationwide cholera epidemic in the 19th century; that occurred in the late summer and autumn of 1853. This was followed by a local outbreak in 1857, mostly confined to the town of Korsør in Sorø County.^9,23,24

Using our method, we identify the 1853 epidemic as excess mortality in Copenhagen city and county starting in July 1853, leading to an estimated 3833 excess deaths over 2 months—a 546% elevation over the baseline mortality of the same period for Copenhagen city. Infant and child mortality was slightly increased during the crisis, but the adult age groups between 15 and 59 years were particularly affected (GM cluster D). This finding is in agreement with reported cholera deaths for Copenhagen discussed in literature.⁹ This age pattern was also observed elsewhere throughout August and September 1853, namely in the counties Aalborg, Aarhus, Hjørring, and Frederiksborg, which is in agreement with well-known epidemics in cities of these counties. Several of the parish registers explicitly mention cholera in the margin of individual burials, substantiating that the excess mortality was caused by cholera.

Excess mortality for all afflicted counties in 1853 amounted to 5154 excess deaths in total. The cholera epidemic in 1857 was confined to Sorø County, with 431 excess deaths. In Figure 4, excess mortality in the affected counties is shown.

Figure 4

Cholera epidemics. Daily excess mortality for some of the counties afflicted in 1853 and 1857. The inset highlights the timing of the 1853 epidemic, with the epidemics in Aarhus and Aalborg counties peaking 3-4 weeks after the epidemic in Copenhagen. Note that the population estimates used here are for entire counties; hence, excess mortality rates for Aarhus, Aalborg, and Sorø counties appear diminished compared with that of Copenhagen city. However, the mortality rates within the cities of these counties were still significant, with previous work showing that almost 5% of the population of Aalborg city perished in the cholera epidemic in 1853.⁹

Scarlet fever

We identified a major mortality crisis in Thisted County between November 1857 and April 1858, with an estimated 659 excess deaths (162% above the baseline). Age-group analysis revealed that 88% of the excess deaths occurred among children aged 1-14 years. In counties Viborg, Aalborg, Randers, Thisted, and Aarhus, mortality crises with a similar age pattern (GM cluster B) were identified during the same period. Four additional mortality crises with different age patterns (GM cluster C) occurred in the same period, with elevated mortality among both children and the elderly, totaling 360 deaths. Although outbreaks of multiple diseases may have occurred simultaneously, we include these additional crises in this group, summing to an estimated 2241 excess deaths. The mortality was also elevated in the nearby Hjørring County, but not significantly enough to constitute a mortality crisis by our criteria.

Severe outbreaks of scarlet fever in the region at the time were discussed in the annual summary of contemporary medical reports.²⁵ The daily mortality in the affected counties is illustrated in Figure 5.

Figure 5

Scarlet fever. Daily excess mortality for 4 of the counties affected by the scarlet fever epidemics in early 1858. The inset displays the excess mortality, smoothed by a 28-day running mean to show timing more clearly. The mortality rate appears to increase slightly above the baseline around the end of 1857, with significant excess mortality in Aarhus County in October and December, Thisted County from January onward, and with peaks in excess mortality a few months later for Aalborg County.

Several mortality crises with similar age patterns were observed in 1859 and 1860 in several counties in southern Denmark (namely, Odense, Svendborg, and Maribo counties). Although these may also have been due to scarlet fever, we cannot distinguish them from other causes of childhood mortality, and so we do not include them in the scarlet fever group.

Pandemic influenza

Our analysis identified simultaneous deadly mortality crises in all counties between December 1891 and March 1892, corresponding with the third wave of the “Russian influenza” pandemic.²⁶ Summing excess mortality, we estimated a total of 6462 excess deaths across the country in this period. The age-specific analysis confirmed the pattern of excess mortality across all ages, with the greatest burden among those 60 years or older (GM cluster F). Our method did not reveal major excess mortality during the first 2 waves in winter 1889/90 and summer 1891. Figure 6 illustrates the 1891/92 pandemic wave.

Figure 6

Pandemic influenza. Excess mortality values for the Copenhagen, Aarhus, and Odense counties are shown. The simultaneous timing in Copenhagen and Odense counties is seen, with a later epidemic observed in Aarhus County. The inset displays the sum of daily excess mortality per 100 000 population for all counties. Although mortality data for the counties in Schleswig-Holstein were in the data set, the data are not included here, because population counts were unavailable.

The 1891/92 epidemics occurred simultaneously across all counties. Similarly, several simultaneous crises between March and May 1900 were identified in almost all counties. Grouped together, the counties affected in 1900 experienced a total of 1874 deaths. The age pattern was comparable to the 1891/92 pandemic, mostly affecting people older than 60 years (GM cluster F). This may suggest that the 1900 epidemics were due to pandemic influenza, as suggested by previous authors.²⁶

Discussion

We studied Danish mortality crises in a 100-year long period (1815-1915), using digitized parish burial registers. We identified 418 crises with at least 50 excess deaths each, totaling around 60 000 excess deaths between 1815 and 1915. These mortality crises were categorized and grouped according to their signature features, including age pattern, sex, duration, and seasonality.

A goal of our work was to develop strategies for identifying the causes of major mortality crises from all-cause mortality data alone. Although Danish parish records do not register cause of death consistently, we were able to refine and validate our findings by consultation of historical sources, determining plausible causes of the deadliest events identified, which account for 50% of the 60 000 excess deaths. Further investigation of contemporary medical reports and other source material could reveal information about the remaining mortality crises.

Our analysis of 19th century excess mortality in Denmark was not exhaustive, and we present findings only on particularly deadly crises. The full table of all 418 mortality crises (Table S2) lists additional crises, some of which are readily identified as epidemics documented elsewhere (eg, the 1837 pandemic influenza).^26,‑28

Our work illustrates how the signature features of mortality crises can guide the characterization of highly deadly epidemics. As an example, the age structure of immunizing childhood epidemic diseases such as scarlet fever and measles is easily distinguished from the older age patterns from cholera and pandemic influenza. The methods for clustering age distributions described in this work were used to guide manual investigations, but future methodological work could expand such age-structure analysis further.

Some of the epidemics discussed in this work are well recorded and have been documented elsewhere based on other sources of data. These include studies on the 1850s cholera epidemics, contemporaneous outbreak studies,²⁴ as well as modern analyses^9,23; studies of the pandemic influenza in 1889-1892 and 1900^26,29; and analysis of the 1826-1832 epidemics.^22,30 This literature describes these epidemics in a degree of detail that cannot be obtained from all-cause mortality data alone. However, because data for rural regions are not easily available, the literature typically focuses on confined geography or larger towns (mostly Copenhagen) and cannot easily investigate the geographic pattern of an epidemic.

Comparing our results with the literature, we observe that some of the referenced studies identify more epidemic deaths than we determined using our method. In particular, Phelps et al.⁹ found 4663 deaths registered as cholera deaths for Copenhagen in 1853, whereas we find 4411 excess deaths for the city and county combined. Although this discrepancy may be due to geographic differences, it highlights that our method only considers excess deaths in periods significantly elevated over the baseline—a baseline which for Copenhagen may include deaths registered as cholera.

Not all mortality crises identified here have received thorough scholarly attention. The identified scarlet fever epidemics of the late 1850s were among the deadliest in Jutland, devasting the childhood population. Our method may help recognize such deadly but underdocumented crises.

Previous studies have illustrated the great value of quantitative analysis of the age structure of mortality, showing that changes in the age distribution follow the severity and timing of disease, both in historical^31,‑33 and contemporary pandemics.^1,34 To analyze the age structure of mortality in this work, we grouped age-specific deaths together into 6 clusters, allowing us to easily distinguish between crises with differing age structures. Simultaneous but unrelated mortality crises may be discerned using this clustering. One example of this was the cholera epidemics, for which a signature feature was the short epidemic period and late summer/early autumn seasonality. Two mortality crises with similar timing were observed in Haderslev County in 1853 and 1857. Despite the similarity in timing, the crises in Haderslev County had very high mortality among children aged 1-15 years, placing them in GM cluster C, whereas the cholera-epidemics were identified as cluster D. Historical sources do not suggest an outbreak of cholera in Haderslev County; however, scanned newspapers reported a dysentery outbreak in Haderslev town and county. This example illustrates the necessity of considering multiple signature features and historical sources; shallow analysis of timing alone could have erroneously suggested that cholera struck Haderslev in 1853 and 1857, despite historical sources suggesting otherwise.

Although the sex ratio was not considered in the results we discussed here, we identified several mortality crises with high male mortality during the first and second Schleswig wars, in the years 1848-1850 and 1864, respectively, in agreement with mortality among soldiers in battles. Worsened living conditions due to the conflicts and disease among traveling soldiers may also have played a role in these crises. Not all mortality crises with high male mortality occurred in counties in which battles took place. In Odense County in 1849, the number of soldiers who died in a field hospital near the town of Middelfart resulted in increased male mortality despite no battles taking place there. Similarly, we identified excess mortality among men aged 15-39 years in Copenhagen in 1864. These deaths occurred far from the war zone; however, parish registers explicitly mention “bullet wounds,” which is consistent with the burial of returned battle-wounded soldiers. Although only a few of the mortality crises identified were war related, this illustrates that our analysis identifies mortality crises in general, and not only epidemics of infectious diseases.

Taken together, all excess mortality we identified only constitutes ~ 1.5% of all deaths in the 100 year period we investigated. At the time, nonepidemic infectious diseases such as tuberculosis, pneumonia, and bronchitis were leading causes of deaths,¹³ and in our study of the medical reports, we found that mentions of deaths due to typhoid fever or dysentery in small villages were common. Similarly, epidemics of diseases with low mortality risk, such as measles or contemporary malaria,³⁵ were also common. Our methodology only captures the deadliest short-term epidemics, in part because of methodological choices about the thresholds for defining mortality crises (see Supplementary Material), and in part because we omit mortality crises with fewer than 50 excess deaths. Omitting mortality crises with fewer than 50 excess deaths could disfavor severe epidemics in counties with a small population, even though the epidemic may have been devastating locally. Further work may identify more local epidemics as well as less deadly epidemics; however, for diseases without epidemic patterns, other methods must be applied. Furthermore, disentangling epidemic mortality from other causes requires careful study of other sources of data, such as death certificates.

To calculate a reliable baseline, we aggregated the parish-level registry data into counties representing the geographic administration at the time. Our analysis thus omits detailed information about individual parishes’ geography and how mortality differed locally between neighboring parishes. A thorough analysis of individual parishes—focusing on parishes with adequate population size and periods with substantial excess mortality—will be the subject of future work. Such work may shed light on the urban and rural dissemination of infectious diseases. Small populations may require different methodologies from those presented here, because statistical fluctuations may be more significant for small populations.

Population sizes were not considered directly in our identification of mortality crises. One reason was to focus on a simple statistical measure based on a single data source (ie, all-cause mortality), in agreement with methods described in literature.¹⁵ Although overdispersion in mortality data may be a concern, most counties were comparable in size. However, population sizes doubled over the 100-year period, which may affect the applicability of our methods throughout the full period studied. Another reason for omitting population sizes was practical: thorough reconstruction of Danish demography over 100 years would be a tremendous task, exceeding the scope of this work.

Historical records are increasingly becoming available in digital form thanks to improvements in automatic transcription tools and the valuable efforts of transcription experts and volunteers. Here, we demonstrate how such data can be used to identify and characterize periods of excess mortality and determine signature features characterizing mortality crises. Our systematic characterization and clustering of age patterns identify mortality crises of interest for further study, which may assist the otherwise difficult and time-consuming task of studying mortality from contemporary sources alone. Our study provides an overview of the most important signature features in the Danish 19th century setting and represents a contribution toward a detailed library of historical mortality crises and their causes. Danish 19th century data can be considered a training set with opportunities for validation and from which to develop strategies for identifying different types of deadly epidemics and pandemics in all-cause mortality data, in data from other countries, both historical and modern.

In future work, we aim to expand the presented methodology by incorporating additional signature features and to extend our work across both time and geography to better understand the dissemination of infectious diseases, both historical and modern.

Acknowledgments

We thank the Danish National Archives for providing us the data used in this work and for being a great source of assistance in discussions about the contents of the data. We are grateful to the Nordemics Consortium for useful discussions about this work. Furthermore, we are thankful to the work of Ancestry and their transcribers. Parts of this work were presented as an oral presentation by R.K.P. (talk no. 0366) at the Epidemics 9 conference, Bologna, Italy, November 30, 2023 and by M.v.W. (paper number 123) during the “NordicEpi 2024” conference, Copenhagen, Denmark, June 14, 2024.

Supplementary material

Supplementary material is available at the American Journal of Epidemiology online.

Funding

This work was supported in part by funding from the Danish National Research Foundation (grant DNRF170), the Carlsberg Foundation (grant CF20-0046), and NordForsk (project 104910).

Data availability

Files containing data aggregated by county level and on 5-year age intervals are available at https://github.com/PandemiXcenter/SignatureFeatures19thCentury.