Introduction

Caregiving in dementia has received considerable research attention due to the enormous demands it poses and the significant resources required to care for someone with cognitive impairment or dementia1,2. For instance, caregivers of persons with dementia often have to manage neuropsychiatric symptoms such as agitation, aggression, and disinhibition that lead to sleep deprivation and abuse among the caregivers3. Moreover, informal caregivers of persons with dementia may experience chronic stress from juggling multiple roles, which puts them at increased risk for cardiovascular diseases, burnout, depression, and anxiety2,4,5.

During the COVID-19 pandemic, informal caregivers of persons with dementia faced unprecedented unique challenges such as social isolation and extended work hours6. In addition to caregiving stress and social expectations of caring for a family member with dementia, they had to navigate through the constantly changing regulations, as well as concerns about viral transmission6. Especially during periods of surging COVID cases, Twitter and other social media platforms (e.g. Facebook and Instagram) have been used extensively by many individuals to express their concerns and opinions7. With its monthly active user base of about 556 million as of January 20238, Twitter’s publicly available data set provides a novel opportunity to enhance qualitative research methods. Unlike traditional data collection methods (e.g. questionnaires and interviews) that can be costly and time-consuming, social media platforms offer a more feasible approach to access real-time information about users’ experiences and opinions7. Furthermore, public health researchers have also increasingly used Twitter data to complement these traditional methods9,10.

An emerging body of literature has used a Twitter-based approach to identify the perceptions of people with dementia and their caregivers, especially in the context of the COVID-19 pandemic. A recent study analyzed 6938 tweets between 17 and 24 March 2020, and highlighted key areas of concerns (e.g. finding support, maintaining human rights, and continuing care etc.) encountered by informal caregivers of persons with dementia during the initial period of COVID-19 pandemic11. The study also showed how Twitter helped to rapidly disseminate vital information to appropriate stakeholders and authorities, as well as exposing the inefficiencies and breaking points in social care systems. Another study examined 58,094 tweets over a one-year period prior to and after the emergence of COVID-19 pandemic, and found increased emotional distress such as depression and helplessness among the caregivers of persons with dementia during the pandemic12. Other related studies have demonstrated the potential of using Twitter-based approach as a tool for informing online self-management intervention design13, categorizing dementia caregiver’s depression levels14, and understanding dementia-related stigmatization15,16,17. This provides some evidence that Twitter may be a viable alternative for obtaining a snapshot of public sentiment.

However, there are two key research gaps in existing literature. First, there is a paucity of studies exploring the potentially changing needs of informal caregivers of persons with dementia over a broader period. Second, few studies have focused on tweets related to individuals’ personal, lived experiences of dementia caregiving. To address these gaps, this qualitative descriptive study harnessed Twitter’s rich, real-time database18, by analysing tweets in English over a 10-year period (i.e. from January 2013 to December 2022). In particular, we sought to identify tweets related to personal, lived experiences of dementia caregiving, and examine any difference in perceived caregiving experience before and after the COVID-19 pandemic. It is hoped that the study findings will uncover the priorities of informal caregivers of persons with dementia and inform relevant stakeholders on how to better support and engage with them.

Methods

Data selection

Tweets posted in English over a 10-year period (1st January 2013 to 31st December 2022) were extracted through Twitter’s Application Programming Interface (API) platform, using an academic developer account that allows the download of all tweets (i.e., not sampling) of up to 10 million tweets per month. Of note, the data were extracted when the academic developer account was still available for use by researchers, prior to the new changes to Twitter’s algorithm and its rebranding to X Corp in April 2023. The relevant tweets were identified using search terms such as “dementia”, “caregiver”, “carer”, “caregiving”, “Alzheimer’s” and “neurocognitive disorder”, with details of the search strategy further shown in Supplementary Material 1.

Non-English tweets were excluded due to the methodology limitation of the machine learning model employed for analysis. Duplicate tweets and retweets were also excluded. Tweets related to “individuals’ personal experiences of dementia caregiving” were selected using a supervised machine learning approach – details on this supervised machine learning, as well as our definition of “personal experiences”, are elaborated in the next paragraph. All data used in this study were collected according to Twitter’s terms of use. Ethics approval for the study was granted by the SingHealth Centralised Institutional Review Board of Singapore (reference number: 2021/2717). Where applicable, R (version 3.6.3) and Python (version 3.7.13) were used for quantitative analyses.

Identifying tweets specific to personal experiences of dementia caregiving

First, a random sample of 500 tweets were selected from the extracted dataset. Two investigators (LCA and TML) independently labelled each tweet as either describing a “personal experience” or “non-personal experience”. Tweets describing “personal experiences” were identified based on two criteria: (1) an individual’s perspective related to dementia (i.e. not organizations that talk about dementia in general, or referring to news on dementia); and (2) an individual’s personal, lived experience related to caring for person(s) with dementia (i.e. not general narratives or observation on dementia caregiving which are not based on an individual’s personal experience). The investigators had a good agreement in labelling the tweets, with a kappa of 0.971 (95% CI: 0.938–1.00; P < 0.001).

Next, the 500 labelled tweets were used to fine-tune a pre-trained Bidirectional Encoder Representations from Transformers (BERT) model19 to identify tweets describing personal experiences related to dementia caregiving. BERT is one of the state-of-the-art approaches in natural language processing, to understand the meaning of words and sentences in a way that is like how humans do, by taking into account the context in which they are used. It was developed by Google in 201819, and has since been deployed in the Google search engine for all the English language search queries20. Unlike previous language models, BERT can process text bidirectionally, i.e., it can consider both the preceding and following words in a sentence when predicting the meaning of a given word. This could be made possible because BERT has been pre-trained on a large volume of text data in a transformer-based neural network architecture, through the approach of masked language modelling which involves randomly masking some words in a sentence and training the model to predict the masked words based on their surrounding context19.

In this study, we adopted the approach of transfer learning, whereby we fine-tuned a pre-existing BERT model21 to allow us to identify tweets describing personal experiences related to dementia caregiving. This is a well-established approach to significantly reduce the amount of data and computational resources needed to achieve high accuracy in text prediction, because we can leverage a pre-trained BERT model that has already learned a rich set of representations of language and adapt the pre-trained model for the specific prediction task that we require19,20. A simplified illustration on the relationship between pre-training and fine-tuning text models is shown in Fig. 1. In the fine-tuning process, we split the 500 labelled tweets into two datasets, with 80% used as a training dataset and 20% used as a test dataset. In the training dataset, the pre-trained BERT model was trained to predict “personal experience related to dementia caregiving”, using 10 epochs as well as class weighting to account for class imbalance in the training dataset. The fine-tuned BERT model was then tested in the test dataset, where it demonstrated 95% accuracy in predicting the correct label. Following the training and testing processes, the fine-tuned BERT model was then used to predict the labels in the full dataset to select the subset of tweets related to personal experience on dementia caregiving.

Unsupervised learning of selected tweets

Thereafter, Topic Modelling22 was used to identify themes in the selected tweets related to personal experiences on dementia caregiving. Topic Modelling is an unsupervised machine learning technique that analyses words and phrases within text documents to extract hidden patterns and group similar concepts together to generate interpretable topics. It is akin to thematic analysis in traditional qualitative methodology; but unlike thematic analysis, Topic Modelling does not require manual labour to group the text data and hence is well-suited for analyses of large volume of text data such as in this study. Specifically in this study, we utilized BERTopic23, a BERT-based Topic Modelling.

Thematic analyses to further refine the output from unsupervised learning

Output from Topic Modelling was examined by two investigators (LCA and TML) to ensure coherence of the identified topics. Descriptive label of each topic was manually crafted by the same investigators based on identified keywords and sample tweets of each topic. Then, the topics were further grouped into three themes by the two investigators who have discussed and iterated before reaching a consensus on the final themes, using the inductive and iterative processes of thematic analysis as introduced by Braun and Clarke24, with investigators: (1) familiarizing themselves with the keywords and sample tweets; (2) producing preliminary codes; (3) formulating overarching themes; (4) reviewing and refining themes; (5) defining and specifying themes; and (6) producing a write-up.

Examine dementia caregiving trends, and any difference related to COVID-19

To identify any shifts in caregiving experience, we further analysed the tweets based on three distinct time periods: 2013–2016, 2017–2019, and 2020–2022. The years 2013 to 2019 were considered as pre-COVID-19 and were compared with post-COVID-19 tweets (1st January 2020 onwards). Changes across time periods were examined using the Pearson’s chi-square test. As a sensitivity analysis, we also stratified the analyses by geographical regions (i.e. North America, Europe, and other/unknown continents) to determine if there was heterogeneity across different regions.

Results

Brief descriptions of the included tweets

A total of 693,458 initial tweets were extracted between the period of January 2013 to December 2022. After excluding non-English tweets, duplicated tweets, and tweets of non-personal experiences on dementia caregiving, 44,527 individual tweets related to personal experiences on dementia caregiving were analysed (Supplementary Material 2). Figure 2 depicts the geographic distribution of tweets, with majority of them originating from North America (37.2%) and Europe (24.5%); the rest came from Australia (1.9%), Asia (1.1%), South America (0.4%), Africa (0.4%), or unknown locations (34.3%).

A total of 10 topics were identified from the tweets after applying topic modelling. These 10 topics constituted 77.7% of the included tweets; the remaining tweets were omitted from the current results as topic modelling generated a miscellaneous topic that grouped all remaining (unfitted) tweets together. Following thematic analysis by the two investigators, the 10 topics could be further grouped into three salient themes that highlighted different key issues in dementia caregiving. The topics and themes are summarized in Table 1, as well as further described in the following paragraphs. A detailed list of representative tweets from each topic, as well as tweets that did not fit well (i.e., grouped under a miscellaneous topic), is available in Supplementary Material 3.

Theme 1: challenges of dementia caregiving

Theme 1 comprised four topics that depicted challenges in caregiving for persons with dementia. Majority of the tweets centered around Topic 1 that describes personal recounts of challenges in dementia caregiving.

"#Carers often face struggles more difficult than expected. The swift progression of my mother’s #dementia was more than this #caregiver was truly prepared for; she went from 2 to 7 on the fast scale in less than 2 years. #braininjury #alzheimers."

"My dad has been in a rougher than usual patch with his health lately, so I’ve been feeling more #caregiving stress. Maybe that’s why this hit me harder than I’d expected. #Alzheimers #dementia #mentalhealth."

Another challenge revolved around concerns during the COVID-19 (Topic 2) pandemic and expressed emotional burden as illustrated by the tweet below.

"I just spent 2 years caring for and then grieving for my MIL with dementia. My household is completely unwilling to risk any covid-related cognitive decline, & I don’t want to pass covid to someone else’s elderly family member. Being a dementia caregiver is brutal."

"Yup, I’m a teacher with 2 school aged kids, so quite likely we’ll be carriers at some point. But I also help care for my father-in-law (with severe Alzheimer’s) and brother-in-law (wheelchair bound). They have carers but rely on us for so much. If the carers get sick, then what?"

Many caregivers described their helplessness and their struggles with financial difficulties (Topic 10).

"I could truly use the help right now. I am a single mom of 2 boys and taking care of my dad who is very sick with dementia, congestive heart failure, a bunch of other health conditions and on oxygen. I haven’t been able to work because I am his full-time caregiver."

"I’m so desperate for help! My car is 2 months behind and im the main caregiver for my dad with dementia and mom w/ cancer and I’m responsible for taking them to appointments and going to work."

Theme 2: strategies to inspire caregivers

This theme comprised three different ways that caregivers might adopt to cope, inspire and connect with people with dementia. First, singing and music were recommended by caregivers as an activity for a person with dementia to improve their wellbeing and as a prompt for reminiscing (Topic 7).

"It’s so good to see my husband remembering the words to songs. He doesn’t speak much now yet he can sing every word.’ A carer of someone living with dementia. Singing has a great effect on those living with dementia. Find a local group #DAW2019."

"I used to be a dementia carer. I saw first-hand how much music and singing meant to the people I cared for. It’s amazing what you and the choir are doing."

Second, caregivers encouraged and supported others by sharing their coping experience through dementia-related books (Topic 8).

"Thank you for writing this beautiful book! I’m now my Mom’s caregiver & she has moderate dementia. Reading this helped!"

"I laughed and cried my way through this brilliant book in a few hours I couldn’t put it down. What a fantastic insight into a very personal tragic situation. Cannot recommend highly enough!! #mymaddad #carer #Alzheimers #parent #love."

Finally, caregivers also shared films that resonated well with them and served as a means to initiate meaningful conversations (Topic 9).

"This is such a compelling short film that gives such an accurate insight into the life of a carer. These women are incredible. Dementia care: It’s not dementia killing me, it’s exhaustion."

"Signed, this film has hit me hard, dad had #Alzheimers and it took it’s toll on mum his main carer, now I’m a carer for my husband with #MND; the similarities are endless - help is needed to help those caring for others to stop them from going under."

Theme 3: dementia-related stigmatization

This theme centered around negative attributions of dementia symptoms, and stigmatizing terms such as “disqualifying someone from a job due to dementia” and “senile” were observed.

"Some medical conditions should automatically disqualify a person from the job. Dementia is one. I’m a caregiver, too. And I’ve seen the drastic changes that happen in 4 years. He will not stay at this stage if this is, really, his diagnosis. It gets way harder fast."

"You really need to rethink saying he is not senile. I have been caregiver to people with dementia for 15 years. Anybody that listens and looks at him without bias knows he has dementia."

Some caregivers used dementia as a means to devalue others, such as Joe Biden (topic 4) and Donald Trump (Topic 5), through the use of pejorative terms like “demented” and “diseased” in tweets that depicted incompetency.

"I am a Caregiver for Dementia patients and Biden def has the beginnings of Dementia and the fact that 81 M people voted this guy in is mind boggling! A demented old man running America, they are the one’s who need their heads examined!"

"Dear Resistors: Take heart. Trump is diseased. He is crumbling. I took care of my mama w/Alzheimer’s, & I see many parallels. Sure, I’m no MD, but as a caregiver on the front lines for many years I can tell U this Trump ain’t making it to 2020."

Temporal trend of the identified themes

Over time, there was a consistent rise in the total proportion of tweets relating to dementia caregiving, as seen in Table 2. The proportion of tweets related to Theme 1 (challenges of caregiving in dementia) remained largely stable and dominated the discussion, accounting for about 91–92% of tweets throughout the study period. However, there was a notable decrease in the proportion of tweets related to Theme 2 (strategies to inspire caregivers) from 6.8% (before 2017) to 2.5–3.2% (2017 onwards). On the other hand, the proportion of tweets related to Theme 3 (dementia-related stigmatisation) demonstrated an increasing trend over the study period. It rose from 1.3% in 2013–2016 to 4.4% in 2017–2019, and further to 6.1% after the pandemic. The results remained consistent in the sensitivity analyses (Table 3), although the decrease in Theme 2 was mainly seen in Europe, while the increase in Theme 3 was more prominent in North America and Other/unknown continents.

Discussion

Summary of findings

Social media enables caregivers from diverse backgrounds to exchange their opinions and experiences in caring for persons with dementia. Thematic analysis of the dementia caregiving tweets revealed three salient themes: (1) challenges of caregiving, (2) strategies to inspire caregivers, and (3) dementia-related stigmatization. Temporal analysis suggested sharing of personal experiences of dementia caregiving was a topic of increasing interest on Twitter, as evidenced by the noticeable rise of relevant tweets from 2013 to 2022. In addition, a comparison of pre- and post-pandemic tweets revealed a notable increase in tweets containing dementia-related stigmatization, especially in North America and in Other/unknown continents (and less so in Europe).

Interpretation of findings

In contrast to prior studies which evaluated tweets on dementia caregiving11,12,13,14,15, our study covered a broader time period with a focus on personal experiences in dementia caregiving. Consistent with past literature1,2,11, our findings showed that the tweets were dominated by the challenges faced by caregivers of persons with dementia (Theme 1), with narratives centered around emotional tolls (Topic 1 and 6), financial hardship (Topic 10), and anxiety over the COVID-19 pandemic (Topic 2). These findings summarize the key concerns experienced by caregivers over the past decade, which further emphasize the lack of caregiver support and the cumulative effects of caregiving on caregiver’s mental and physical health. For example, caregiver stress and anxiety related to several aspects of caregiving including finances may place a strain on the relationship with the care recipient and other family members25, as well as affect the quality of care provided to the care recipient, particularly in times of pandemic crisis26.

Unsurprisingly, the percentage of tweets discussing challenges in caregiving remained constant and continued to be the main topic of conversation post-pandemic. This suggests that the challenges faced by caregivers remained a significant concern even after the pandemic, highlighting the need for continued support and resources for caregivers. The findings on strategies to inspire caregivers (Theme 2), such as connecting with persons with dementia through music (Topic 7) and inspiring others through books and films (Topic 8 and 9), are less expected but noteworthy in buffering the impacts of caregiving stressors. Even as these narratives to inspire caregivers represent a relatively smaller proportion of tweets, the presence of such tweets provide some semblance of caregiver self-efficacy and personal mastery that is an important protective factor against psychological stressors of caregiving27.

A small percentage of tweets centered around dementia-related stigmatization (Theme 3), which is not unexpected and contributes to the growing body of research on stigmatization towards dementia in social media platforms15,16,17. However, a troubling and unexpected finding is the internalization of public stigma among the caregivers and the application of it as a means to devalue others. In general, two types of stigmas have been described in the literature – public stigma refers to the negative perception and stereotypes held by the general public; while self-stigma refers to the resulting process of a person’s internalization and agreement of the stereotypes28,29. Arguably, it may still be understandable for members of the public to hold some stigmatizing view towards dementia when they have had limited exposure to the condition. However, it can be concerning when caregivers themselves started to internalize these publicly held stigma and believed in the stereotypes, and to make matter worse, even project these stereotypes externally as a means to devalue others30.

Equally concerning, is the rising trend of dementia-related stigmatization after COVID-19 pandemic, especially in North America and in Other continents. This worrying change could possibly be related to the fact that the US has had two older presidents consecutively, as reflected by the pejorative comments towards Joe Biden and Donald Trump in this theme (i.e. Topic 4 and 5 of Theme 3). The rising trend possibly also coincided with the COVID-19 lockdown around year 2020. Conceivably, the COVID-19 pandemic and subsequent lockdowns have led to a significant shift in communication methods. With restrictions on social activities, many may have turned to social media to stay connected and to maintain support networks. This rising trend of dementia-related stigmatization can be concerning, because repeated exposure to such (mis)information may exert a lingering influence on people’s judgement even after correction31,32,33. This is known as the continued influence effect, which can impact stereotypes and pose as a major barrier to mitigate dementia-related stigmatization. If left uncurbed, these negative stereotypes and stigmatization may have detrimental consequences. For example, self-stigma in caregivers may result in lowered self-esteem34, and may distance themselves from social contacts due to fear of embarrassment or being judged by others27,35,36. This could potentially lead to a delay in seeking help and intensify the burden of caregiving37,38. Collectively, our findings make a case for greater discussion on the nuances of stigmatization in dementia, shed light on the public perception of dementia over the decade, and highlight the pressing need to counteract the spread of dementia-related stigmatization in social media.

Implications of findings

Arguably, our findings may possibly reflect a new generation of caregivers who are more adept in using technology, whereby social media could be used as an avenue in the future to survey the prevailing concerns and challenges related to caregiving in dementia. Social media could be explored to identify prevailing challenges, which in turn allow the deployment of additional community resources to better support the caregivers in real-time. For example, narratives on the topic of contracting COVID and increased vulnerability among people with dementia (Topic 2) exhibit a real fear that caregivers faced, which have also been highlighted in the literature39. Another example is the mental health strain (topic 6 and topic 10), where caregivers experienced significant emotional impacts due to COVID-19 restrictions which have led to reduced support services and feelings of exhaustion and helplessness. Additionally, caregivers often face dementia-related stigmatization (Theme 3), contending with societal perceptions of themselves and their loved ones with dementia40. The presence and growing trends may help point policymakers and healthcare professionals towards areas to focus on in public health campaigns. For instance, policymakers could consider using social media platforms to educate the public about dementia and its impacts; sharing positive and realistic stories of people living with dementia and their caregivers; and reporting or flagging dementia-related content that is stigmatizing15,16.

While there is an increasing number of caregivers who take to social media to inspire other fellow caregivers (Theme 2), such tweets are proportionally low compared to those that emphasize on the negative aspects of caregiving (Theme 1). Perhaps, policymakers and Alzheimer’s Associations around the world could tap on positive role models (i.e., caregivers who have successfully navigated through some of the key challenges) to share their experience on social media thus leveraging social media in reaching out to a larger group of caregivers to build their resilience and coping strategies in caregiving. This approach is also consistent with the Social Learning Theory, which posits that caregivers may learn better – especially with regards to caregiving skills as well as self-efficacy in managing caregiving stress – when they can observe and model people around them who have had similar experience to the caregivers themselves41,42,43.

Limitations

Despite the strength of Twitter-based research, several limitations are worthy of further attention. First, Twitter users may not represent the broader population, as not all caregivers of persons with dementia have a Twitter account, and a significant portion of Twitter’s demographic may be individuals in their twenties and thirties. Second, Twitter may be susceptible to personal perspectives and bias that could lead to under- or over-representation of biased opinions. Furthermore, individuals who posted on Twitter are often more likely to be well-educated and receptive of using social media applications. Third, our analysis was limited to English-language tweets due to constraints in our machine learning model methodology and human interpretation capabilities. This may not fully represent the global population. Although there was a rising number of tweets relating to dementia caregiving over time, without access to data on the overall volume of tweets on the Twitter platform, we cannot ascertain whether the observed trends reflect changes in the broader Twitter landscape or specific shifts in the prevalence of dementia caregiving-related discussions. Fourth, only publicly available tweets were extracted, and there is a possibility of information exchange in private account exhibiting different thematic patterns. Fifth, the implementation of diverse quarantine policies across different regions and timeframes may have variably affected individuals with dementia and their caregivers. Furthermore, majority of tweets analysed in this study came from North America and Europe, which would affect the generalizability of the results. Hence, our results should be interpreted with caution, considering the potential impact of these policy disparities and the heterogeneity across the regions on the observed outcomes. Sixth, although topic modelling is able to process large number of tweets, this method might not be as in depth as traditional qualitative methodology. In particular, topic modelling identifies similar topics by looking for pattern in the tweets, which may result in some tweets being clustered together due to use of common phrases. For example, Topic 4 and Topic 5 seemed to group tweets which centered around “Joe Biden” and “Donald Trump” respectively. Which this approach may not be wrong, it may not capture nuances in meaning of the different tweets. Finally, due to the brevity of tweets, it is possible that tweets which are related to caregiving, but devoid of the relevant search terms, may have been inadvertently excluded from the analysis. Future research could consider expanding the search by including other languages and other social media channels to ensure a comprehensive reservoir of information pertaining to caregiver concerns.

Conclusion

This study sheds light on the increased engagement of Twitter platform by caregivers of persons with dementia and is among the few in current literature to examine the salient concerns among caregivers over the past decade. The COVID-19 pandemic has significantly increased the stress and emotional burden on caregivers. This is evident from the prevalence of tweets highlighting the negative aspects of caregiving, such as financial and emotional challenges, which can strain relationships and consequently affect the quality of care provided. Strategies to inspire caregivers through music, books and films are also uncovered, which are important coping strategies to shield against psychological stressors. Notably, although only a small percentage of tweets indicate dementia-related stigmatization, the significant increase in North America and Other continents is concerning. This rise in stigmatizing tweets post-pandemic among caregivers reflects a worrying trend in caregiver’s perception of dementia on social media. Addressing the issue of self-stigma among caregivers is of paramount importance, especially when it manifests in the form of demeaning others. Taken together, these findings highlight the potential utilities of social media in the context of dementia caregiving – to gain insight into pressing challenges encountered by caregivers; as an alternative avenue to empower caregivers in sustaining their role; and as a means to counteract the spread of dementia-related stigmatization, which if left unchecked, may further debilitate caregivers.

Fig. 1
figure 1

An illustration of the Bidirectional Encoder Representations from Transformers (BERT) based large learning model (LLM) to the prediction of text.

Full size image
Fig. 2
figure 2

The geographic spread of Twitter feeds concerning caregiving experiences in dementia. Each tweet is indicated by a black dot on the map. Figure created in R (version 3.6.3) ggplot2 package.

Full size image
Table 1 Themes and topics related to dementia caregiving (N = 44,527).
Full size table
Table 2 Temporal trends of dementia caregiving themes between 2013 and 2022 (N = 34,576).
Full size table
Table 3 Temporal distribution of dementia caregiving themes across North America, Europe, and Other/unknown continents between 2013 and 2022.
Full size table