First, we identified all-time film deaths using the Cinemorgue Wiki [24] which is a crowd-sourced database that allows any members of Fandom to update death by cause by film and film year (1895‐2023), with a short description of the death. The site is organized by death type or by actor (not by character) and each entry is a short 1‐2 sentence description of the film death. It is inclusive of any movies publicly entered, and thus not limited to American film deaths only. Conveniently, Cinemorgue has historically provided an .xml download of their entire website (as of August 16, 2023) and other researchers have previously classified cinematic deaths using Federal Bureau of Investigation (FBI) homicide specifications [25]. Refer to Table 1 for a basic description of the database entries.

Using this existing script, we adjusted the classifying portion to identify deaths according to the CDC NVSS’s top mortality causes by year for 2021 [1]. Our script parsed the description of each film and reduced it to word stems by film entry. We used generally broad word stems that captured the colloquial terms describing different mortality causes (eg, “heart attack” and “heart failure”). We then categorized film deaths by gender using a reasonably comprehensive and inclusive classifier from PyPi project gender-guesser, which was trained on a representative international database of names that were independently validated by native speakers of many languages [26]. When applying gender-guesser in this study, all 40,000 plus names were included to account for various languages and ethnic backgrounds of actors represented in film (the Cinemorgue Wiki includes movies from many different countries). Furthermore, the code allowed for the classification of mostly male, mostly female, androgynous, and unknown names. Table 2 enumerates the initial name classification.

Based on a random manual review of classification, “mostly male” and “mostly female” seemed to accurately categorize well-known actors, “mostly male” was grouped with male, and “mostly female” was grouped with female. Based on the size of the data and this classifier, we removed androgynous or unknown names (collectively referred to here as “other”), discussed in the Limitations section. The 2021 top 10 leading causes of death in the United States are depicted in Table 3 [1]. This list actually comprises 13 leading causes of death because the ninth and tenth leading causes of death by gender differ, and a category of “other” is including to reflect the remaining causes. If absent from the official report, these data were queried directly from the CDC WONDER database [27] using the same International Classification of Diseases, Tenth Revision (ICD-10) parameters from the NVSS [1]. We calculated the ratio of mortality by gender by performing a simple ratio of male or female deaths by mortality cause over total deaths by that cause. It is important to note that the top 10 leading mortality causes differ year-to-year, and 2021 was the second year to include COVID-19 deaths. In addition, the medical terms describing these causes of death are extensive and often highly technical; for ease of understanding, we provide additional notations to specify the colloquial terms for these diseases.

In both datasets, we calculated the ratio of mortality by gender by cause by performing a simple ratio of male death (or female death) by mortality cause over total deaths by that cause. Thus, the sum of all ratios should equal 1. From the ratios of mortality by gender in film and NVSS data, we constructed a bar graph visualization. Then, we explored the significance of the difference in gender ratio by cause of death between film and real life using the Pearson chi-square test [28,29].

It is important to underscore how to interpret significant P values in this context [30-32]. Accordingly, P values <.5 do not “measure the probability that the studied hypothesis is true, or the probability that the data were produced by random chance alone”, and are merely “one approach to summarizing the incompatibility between a particular set of data and a proposed model for the data” [30]. That is, if there is a significant P value for the chi-square test for any given cause of mortality, this means that the data seem to suggest there is a significant difference between expected and observed frequencies of causes of death by gender between film and real-life but does not establish causality.

All US mortality data used for this study are publicly available, anonymous, and consolidated at the national level. All data have been deidentified. Cinemorgue data comprise fictional, in-film deaths. This study is exempt from ethical approval, as the research involves the analysis of large-scale, publicly available datasets that are legally accessible to the public and do not require any special access or permissions. The data do not include any information through which individuals can be readily identified, either directly or indirectly. Because the data are both publicly available and do not involve interaction or intervention with human participants, nor do they involve the collection of private, identifiable information, this study qualifies for exemption from institutional review board oversight.

Film and real-life mortality differ significantly for some, but not all, leading causes of death. As expected, film is not representative of real life even setting aside gender considerations. From a practical perspective, this makes sense in the context that high-tempo, salacious, or dramatic deaths are likely far more interesting to an audience than everyday causes of mortality. In film, for example, the proportion of “other” deaths represented far outpaces the actual leading cause of death (diseases of the heart) at a factor of over 800. When considering gender, there are significant differences in the ratio of represented death for accidents, cerebrovascular disease, diseases of the heart, suicide (intentional self-harm) and all other deaths not captured using the top 10 leading causes of death classifier, among sufficiently large samples within the Cinemorgue Wiki database.

While our results do not indicate causality, they provide insight to the representation of mortality in popular media. From visualizations and statistical comparison of ratios, there is meaningful descriptive power in the ratios of film versus real-life representation of mortality by gender. In view of recent efforts by the American Heart Association to provide better equality of cardiovascular care for women [3] as well as general research regarding treatment differences by gender [33], the underrepresentation of film deaths by gender demonstrate that death is also lacking in equitable representation on screen. As there is a general consensus in the literature that stereotypes can be exceedingly harmful and contribute to sexism, harassment, and violence against women [13] as well as other gender identities [34], it is reasonable to postulate that a lack of representation in film could exacerbate the public’s misunderstanding of disease burden and processes, and could hinder the recognition of gender-specific health issues among not only the public but potentially among health care professionals.

By cause of mortality, some interesting patterns emerge. For the category of “other,” men are overrepresented in the Cinemorgue Wiki, which is consistent with their overrepresentation writ large in the Wiki (regardless of cause of death). This could indicate that men are represented more often on screen generally, and represented more often in violent deaths, as previously discussed. For diseases of the heart, men are seemingly overrepresented in film, which agrees with the American Heart Association’s recent findings on underrepresentation and lack of understanding of women experiencing heart disease [3]. Likewise, men are overrepresented for death by cerebrovascular disease. However, suicide and accidents are overrepresented for women in film. Based on the suggestibility and risk associated with suicide representations [14], this pattern indicates that future follow-up is needed to understand the relationship between representation and acts of suicide in women.

Limitations of this study center on the classification and cleaning of the Cinemorgue film death dataset, as well as the data itself. First, we used only film deaths in this study, on the assumption that films generally have wider ranges of audiences (not confined to certain streaming platforms) and have a similar format. However, the Cinemorgue Wiki database includes all manners of video-represented deaths, including television and video games, which were excluded in this study based on the assumption that video game deaths may overrepresent homicides, due to the violent nature of many video games. Furthermore, we compare the Cinemorgue Wiki database all-time counts with single point-in-time (2021 NVSS) mortality counts; considering that movies are not only watched in the year in which they are filmed, we consider this a reasonable assumption even though movie popularity changes over time. In addition, in Table 1, we report that 50% (34,268/68,535) of the film deaths in our dataset are from 1995 and later, with 25% (17,132/68,535) occurring between 2010 and 2023, showing that the preponderance of films occur relatively close to the point-in-time mortality counts with many being topically contemporary to the year in which they were filmed.

In addition, our removal of androgynous names may have skewed the resultant data; as illustrated previously in Table 2, 13% of data are effectively “thrown out” of the analysis because it was not classifiable as “female” or “mostly female” or “male” or “mostly male”. While using binary categories is consistent with the CDC’s current reporting causes by mortality, reducing gender to solely men and women does not capture the diversity of gender identities. Considering the connection between media representation and suicide, as well as the significantly higher instances of suicidal ideation in transgender and nonbinary youth than cisgender youth [35], indicates that further research beyond this study is important for understanding the representation of health outcomes for transgender and nonbinary individuals. In terms of classifying cause of mortality in the Cinemorgue Wiki database, the search terms used on the stemmed text were specific but very broad in order to prevent an undercount of deaths, which could have skewed resultant counts. Intrinsically, the Cinemorgue Wiki database is also subject to the normal risks of crowd-sourced data. Members of the public that are heterogeneous in knowledge, skill, and geographical location comprise the crowd [36] that generates the entries in the Cinemorgue Wiki; biases in any of these characteristics could contribute to a geographically skewed or low-fidelity dataset. While CDC data from the United States were used to calculate gender proportions, every film death in the database was included, even if the film was not English-language. Accordingly, while some films may better represent certain realities than others, research has shown that even explicitly far-fetched or out-of-reality story lines can influence beliefs via transportation [37], so we have retained all films regardless of location of origin or setting.

Future research should further detail the corpus of cinematic mortality, and incorporate additional features of each film (eg, age of actors, setting, location, audience scores, and critic reviews) from film encyclopedias and rating websites. This will further elucidate the nature of mortality representation in film relative to viewership. In addition, future work should examine mortality representation across all forms of media, including books, songs, and social media. This would provide a more global perspective on the impact of media—writ large—and capture populations that may not interact with cinematic media. With enough data, the Granger causality test—used to forecast time series from other time series [38]—could be used to understand whether media representation of death by gender lags or leads actual causes of mortality by gender. This would contribute to a better understanding of the interplay between media representation and health outcomes over time. Relatedly, visualizations like a Sankey flow diagram—which provides visualization of time-series-type data in lieu of a formal time series analysis [39]—could provide further opportunities for interpretation of how gender representation has morphed over time. Randomized controlled trials centering on these ideas (perhaps priming viewers with biased images of disease before surveying them on perceived prevalence of disease by gender) could provide support for the case of causality between representation and understanding of society’s disease burden.

Our study compared mortality causes by gender in film and real life using Cinemorgue Wiki data on all-time film deaths and NVSS mortality by cause data from 2021 in the United States. Based on the comparison of gender ratios by cause of death, we identified significant differences in representation for accidents, cerebrovascular disease, diseases of the heart, and suicide (intentional self-harm). Considering current struggles in appropriately managing diagnosis and treatment of disease by gender, highlighting how representation of mortality is skewed by gender in films may encourage a more equitable culture surrounding gender and health. Future work should expand this query to include other forms of media to understand the interplay of representation and mortality.