Australian Diabetes News Media Coverage

The public perception of diabetes is shaped by the news media, which have been widely recognised as an influential source of information related to health.^1-4 How the news media cover health in turn impacts on public policy agendas, policy makers, and the policy process.^3-5 It is therefore important to have an awareness of how news media cover diabetes. Previous studies have investigated diabetes news coverage in North America,^5-7 the UK,^8-10 and New Zealand.¹¹ In this study, we focus particularly on Australia, where approximately 1.7 million people have diabetes, and a further 2 million are estimated to be at high risk of developing type 2 diabetes (T2D).¹² Prior studies of Australian newspapers^13-15 have investigated topics, amount and distribution of diabetes coverage, language use, references to Aboriginal and Torres Strait Islander people(s) and issues, and whether reporting reflects ‘real’ statistics. In this study we target three as yet unexamined research questions, namely:

• Is the reported research based on animal or human studies?
• What news actors and sources appear in the news media?
• Is there any evidence of weight and lifestyle stigma or blame?

Why these particular issues? Firstly, all three issues are associated with ethical and social justice questions. Animal research is the subject of much ethical debate and clearly of public interest. To give a recent example, when three baboons escaped from a Sydney hospital in February 2020, this resulted in considerable media coverage as well as demands for greater transparency in animal research.¹⁶ Concerns have also been raised about the relevance of animal studies to humans and whether findings from such studies can be directly applied to humans. In addition, animal studies are subject to legal acts (e.g. animal welfare acts) and some journals, including the Australian Diabetes Educator, do not publish studies involving animals. The next question – what news actors and sources appear in the news media – relates directly to critical concerns about social representation: who is visible, who sets the agenda, and who is given a voice? The final issue (weight/lifestyle) concerns the negative social consequences and inequalities for people living with diabetes when they are stigmatised or blamed for their condition.

Secondly, this study asks questions that stakeholders in diabetes news might be interested in, i.e. involving an element of (informal and indirect) research co-design. The specific ‘stakeholders’ considered in this study include diabetes researchers, newspapers/journalists, people living with diabetes, and animals (represented by animal rights groups). Initial conversations with diabetes researchers of the University of Sydney’s Charles Perkins Institute identified an overall strong interest in the presence of stigma/blame, with one researcher also mentioning their interest in animal/human studies. In addition, newspaper guidelines¹⁷ and social media accounts (@justsaysinmice) both highlight that research based on animal studies must be treated with caution and reported on appropriately. Animal rights groups, of course, campaign against animal experimentation. In relation to stigma/blame, this is not only of interest to diabetes researchers, it also directly affects people living with diabetes who report diabetes-related stigma¹⁸ and is clearly of interest to diabetes educators and not-for-profit diabetes organisations.

Thirdly, with the exception of weight/lifestyle, these questions remain largely unexplored in research on either international or Australian diabetes news coverage. McNaughton’s¹⁹ study of 81 Australian documents (including 42 media stories) between 1988-2012 showed that overweight and obesity have become cultural signifiers of T2D despite the complex etiology of the condition. Including analysis of weight/lifestyle allows us to investigate whether these results hold for a more recent and much larger dataset (almost 700 media articles) and how they compare to international research, where weight/lifestyle is incorporated in relevant framing or content analysis studies. For example, a 2005 analysis of US magazines and Canadian newspapers published in the 1990s/early 2000’s identified mentions of modern lifestyles as cause of T2D,⁶ while a 2009 study found that obesity or weight gain were most commonly mentioned as cause of T2D in in almost 70% of articles published in US newspapers in 2005-2006, with 40% of articles identifying health behaviour such as diet and exercise.⁵ Focusing just on the New York Times, Stefanik-Sidener⁷ found that the behavioural frame (associated with individual, personal behaviour such as lifestyle, dieting, etc) was dominant in about a third of relevant articles published in the first decade of the 21^st century. In UK news, obesity became central to diabetes coverage, including mentions as risk factor or cause, while lifestyle elements were also associated with diabetes risk, and critiques of public health solutions recommended responses focusing on lifestyle and weight.^{8, 9} Examining five national newspapers in the UK, Hellyer & Haddock-Fraser¹⁰ found that ‘healthy diet’, ‘active lifestyle’ and ‘healthy weight’ were the three most widely reported prevention variables for T2D. Most recently, Gounder & Ameer^11(p8) showed that the ‘Behavioral frame dominates solution attributions of all diabetes types’ in NZ news coverage published in 2013-2014. This means that it is individuals and their behaviour (e.g. dieting, exercising, lifestyle changes) that are most commonly presented as responsible for the solution for diabetes occurrence and management.

Methods

The dataset consists of 694 items from 12 major Australian newspapers (2013-2017) archived in the Factiva database (577 news items and 117 ‘non-news’ items such as opinion, analysis and profiles), where any word starting with diabet– appears in the headline or lead paragraph (for details of the dataset construction, see (20)). While the dataset does not cover 2018-2020 and therefore does not inform us about the impact of COVID-19 on diabetes news media coverage, it can act as a useful baseline for future comparisons. Given the large number of articles in our dataset and the different questions we were interested in, we decided to combine manual and computer-assisted text analysis. More precisely, the techniques applied in this study involved analysis of word frequency and word usage through the software WordSmith²¹ as well as analysis of named entities using SpaCy²² and Stanford CoreNLP NER models.²³ Briefly, WordSmith allows for quantitative and qualitative analysis of words and phrases. The software retrieves every instance of all words in a dataset and presents quantitative information such as the frequency of each word and the number of texts it occurs in. An inbuilt ‘concordancer’ tool provides the researcher with every sentence in which a given search term occurs, facilitating follow-up qualitative analysis (for example, whether the word human is used with reference to human trials). Alphabetic sorting facilitates such analysis (see Figure 1).

Figure 1 Twenty concordance lines for human, sorted according to the surrounding text to the right of the search term

Named entity recognition software automatically extracts entities from a dataset by identifying proper name mentions in text (e.g. groups, organisations), based on pre-trained models. Both techniques are used in this study, but not in relation to all three research questions listed in the Introduction. In all cases, automated quantitative analysis was the starting point for manual qualitative analysis, which was undertaken by Carr in extensive consultation with Bednarek. The specific methods along with the results are introduced below. Further methodological details and coding manuals used are provided via the project page on the Open Science Framework (https://osf.io/jrhx2/).

Results

Animal/human research

Our first task was to identify how much reported research is based on animal or human studies. To do so, WordSmith was used to identify all texts that contain experiment* or trial* or study* or studies or studied (947 total instances in 324 texts; the asterisk * stands for zero or more characters, meaning this search retrieves all relevant forms of the words). Each of these 324 texts was then manually coded for animal/human research (for study details, see https://osf.io/jrhx2/): 164 texts involved studies with humans; 38 involved results based on animal testing, seven involved results based on both; six involved studies where it is unclear whether they are based on animals or humans, and 109 texts are N/A (for example, because they only contain mentions in passing or references to future studies not yet undertaken). We also noticed that articles reporting animal studies often include references to humans or human trials. To study this phenomenon more systematically, we used the concordancer to identify all instances of human or humans or human’s in the dataset. Qualitative analysis showed that when these word forms are used in articles on animal studies, they typically make comparisons to human bodies, refer to future studies with human trials, or explain why human studies are difficult. Arguably, such representation makes the results from animal studies explicitly relevant to humans, and thereby also implicitly justifies them. Illustrative examples are presented in Figure 2.

Figure 2 Ten concordance lines for human(s)

News actors and sources

Our second task – analysis of news actors and sources – is important for identifying who sets the agenda in health news. First, to discover who is mentioned in diabetes coverage, named entity recognition software was used to identify the national/regional groups and organisations mentioned in ten or more texts in the dataset:

• National/regional groups: Aboriginal, Australians, Australian, British, Chinese, Danish, Tasmanian, Tasmanians, Torres Strait Islander, Victorian, West Australians
• Organisations: Apple, Austin Hospital, Australian Diabetes Society, Australian Medical Association, Baker IDI Heart and Diabetes Institute, CSIRO, Deakin University, Diabetes Australia, Diabetes WA, Diabetes Research WA, Federal Government, government, JDRF [Juvenile Diabetes Research Foundation], Juvenile Diabetes Research Foundation, Monash University, National Health and Medical Research Council, PMH [Princess Margaret Hospital], Princess Margaret Hospital, Royal Adelaide Hospital, state government, Telethon Type 1 Diabetes Family Centre, University of Adelaide, University of Melbourne, University of Sydney, St Vincent

(ordered alphabetically, those distributed across at least 20 texts in bold; for methodological details, see https://osf.io/jrhx2/)

Together, the results from this analysis suggest that there is a focus on Australia in the dataset, and that diabetes organisations, universities, hospitals, and government organisations play a big role in shaping diabetes coverage. This is in line with the news values of Proximity (geographical and cultural nearness to the target audience) and Eliteness (the elite status of sources and news actors), which influence ‘newsworthiness’.²⁴

However, this technique does not tell us whether these entities are just mentioned in news articles or whether they act as sources, i.e. are cited by journalists. To find out if this is the case, WordSmith was used to identify the four most frequent reporting expressions in the dataset (said, says, according to and say) and then manually classified the sources that occur with these expressions (pronouns excluded). The coding manual (https://osf.io/jrhx2/) provides definitions and examples for each category and explains our approach to double codings, which were largely avoided. Results are presented in Table 1 as total frequency of codings, percentage of all codings (in brackets), and number of texts.

Table 1 Frequency, percentage and number of texts for each source category

Table 1 shows that the most common cited sources are research findings and announcements; medical and health experts; health advocacy groups; ‘lay’ people; and politicians, government officials and government initiatives. These account for around 88% of all sources, and are the same five categories identified by a previous study on kidney disease.⁴ This shows that these source types are common in Australian health news coverage more generally, again in line with the news value of Eliteness – with ‘lay’ people (individuals who do not speak in any official/institutional capacity, including people with diabetes) only comprising ~15% of codings.

Weight and lifestyle stigma or blame

The final task was to identify whether references to weight and lifestyle are common in our dataset. Both authors independently inspected the 200 most frequent word forms identified by WordSmith, isolating any that could point to cases where diabetes is linked to a person’s weight or lifestyle (dieting, exercising, sleep etc.). The relevant words identified by both study authors as potentially relevant (100% agreement) occurred in 13-27% of the dataset (f = frequency, n = number of texts; % = distribution):

• diet (f 469; n 187; 27%)
• weight (f 430; n 182; 26%)
• obesity (f 256; n 129; 19%)
• exercise (f 252; n 154; 22%)
• food (f 231; n 133; 19%)
• fat (f 208; n 88; 13%)
• lifestyle (f 187; n 126; 18%)
• eating (f 183; n 121; 17 %)
• overweight (f 183; n 103; 15%)

Statistical analysis showed that (except for food) these words are strongly associated with the word diabetes, occurring in its immediate surrounding text. A subsequent qualitative analysis was undertaken of instances where these words (except food) occur within five words to the left or right of diabetes: In 66 of 230 relevant concordance lines, the word co-occurrence linked diabetes and lifestyle or weight in the same sentence, while in a further 151 instances this link was made elsewhere in the same article (coding details in https://osf.io/jrhx2/). These results reflect the prevalent and naturalised link between overweight/obesity and diabetes in Australian popular, academic, and public health discourses since 1998.¹⁹ But they are also in broad alignment with studies of diabetes in print news from North America,^5-7 UK^8-10 and New Zealand.¹¹

Discussion

This study used computer-assisted text analysis to examine critical issues in news coverage about diabetes. Alongside the use of computer software for automated quantitative analysis, the study involved a significant amount of manual analysis: manual coding of over 300 texts (whether they involved animal/human studies); manual analysis of concordances (e.g. 230 lines in the weight/lifestyle analysis), and manual coding of sources (almost 2000 instances).

The analysis showed that articles report research based on human studies more than they report research based on animal studies. However, when animal studies are reported they are often associated with humans or human studies, which may work to discursively justify animal experimentation.

Our analysis of news actors and sources showed that newspapers frequently mention diabetes organisations, universities, hospitals, and government organisations. They cite medical experts most, followed by health advocacy groups. From a critical perspective, lay people – including people with diabetes – are perhaps under-represented, although they are not entirely absent. The analysis of news actors/sources also supports the importance of news values such as Eliteness and Proximity in health news. The results therefore highlight the influence of newsworthiness on health information as presented in the news media.^{4, 5}

Finally, through analysing references to weight and lifestyle in a recent newspaper dataset, the study also identified the need for continued education about the complex etiology of diabetes. It is worth noting here that references to weight and lifestyle are associated with ‘individualised frames’, where individuals rather than structural factors are seen as causing diabetes, and are ‘tied to the perpetuation of blame and stigma attribution’.^11(p3) This can also lead to public perception of the individual as responsible for addressing diabetes and thus to scepticism of health policy initiatives.^{5, 25} Poor lifestyle choices are also emphasised by the food and marketing industry.^11(p4) Future research needs to investigate how weight and lifestyle are linked specifically to (types of) diabetes (e.g. as risk factor, cause, prevention or management strategy), and to investigate more systematically who/what is constructed as responsible for diabetes occurrence and management (e.g. through analysis of frames). Given the declining readership of print news, additional data also needs to be considered, such as other types of media, including social media as well as publications produced by not-for-profit organisations.²⁶