This study did not require institutional review board approval because we used publicly available social media data that do not involve human subjects and do not fall within the scope of Human Subjects Research. Seattle Children’s Hospital’s Institutional Review Board uses the US Department of Health and Human Services (DHHS) definition of Human Subjects Research. Human Subjects Research under DHHS regulations is defined as the investigator obtaining information through intervention or interaction with living individuals; or obtaining, using, studying, analyzing, or generating identifiable private information. Our research is Nonhuman Subjects research according to the Seattle Children’s HRP-101 Human Research Protection Program Plan, P. 3, which uses the DHHS definition of Human Subjects Research [25,26]. The publicly available social media data reported in this paper have been anonymized and contains no IDs, user names, or nonparaphrased tweets.

In this study, we examined pediatric cancer–related communication on Twitter encompassing a representative timeframe ranging from before to after the COVID-19 pandemic. We obtained a total of 182,628 publicly available global tweets from December 11, 2019, to May 7, 2022, using the Twitter application programming interface. An example of the query and timeline information using “#teenagecancer” is available in Multimedia Appendix 1. For this study, we restricted our collection to English-only tweets. We identified a list of hashtags commonly associated with pediatric cancer: #childhoodcancer, #childhoodcancerawareness, #childhoodcancerday, #internationalchildhoodcancerday, #kidsgetcancertoo, #pediatriccancer, #pediatriconcology, #teenagecancer. These 8 keywords were selected because they were representative of hashtags frequently used for pediatric cancer. The prepandemic period was designated as December 2019 to February 2020. The pandemic time period was designated as March 2020 to June 2020. Lockdown and mandatory stay-at-home orders were issued in 42 US states and territories across the United States between March and May 2020 during the height of the pandemic [27]. The postpandemic time period was designated as July 2020 to May 2022 after mandatory stay-at-home orders were lifted in all states across the United States. Removing duplicates resulted in a total of 161,135 tweets from 40,289 unique users. All unique tweets were used for lexicon-based sentiment analysis. Among the 161,135 tweets from 40,289 unique accounts, we then randomly sampled a subset of 800 tweets and analyzed them using a directed content approach. Of the subset of tweets, 300 were randomly sampled and proportionately stratified by pandemic period (prepandemic, during the pandemic, and postpandemic).

We explored the background literature related to cancer and other health conditions to identify a codebook based on directed content analysis for our project [13,33-36]. We identified Sutton et al [13] for lung cancer messages as the codebook that was most relevant and related to our sample of pediatric cancer tweets. We conducted a directed content analysis [37] using codes and coding definitions from Sutton et al’s codebook. Further, 2 of the authors (NL and AO) coded tweets in sets of 10 to iteratively refine and adapt the Sutton et al codebook and definitions to correspond to pediatric cancer-related tweets. We expanded the preliminary codebook to include the emergence of 7 new coding categories that did not exist in the original codebook. We also added a not enough information coding category for tweets that were ambiguous and could not be coded. Furthermore, 2 of the authors (NL and AO) met weekly to discuss and address codebook discrepancies, and further refine the codebook. The entire authorship team met to discuss codebook development and refinement until it was stable and finalized. The same 2 coders (NL and AO) used the final codebook of content of tweets (adapted from Sutton et al [13]; Textbox 1) to independently double-code all 800 of the randomly sampled tweets.

Interrater reliability was calculated as the percent agreement between raters before consensus meetings. Consensus conversations occurred weekly and we referenced the codebook to resolve any discrepancies. The qualitative data were analyzed using DeDoose (Sociocultural Research Consultants, LLC) software for code frequency counts and code co-occurrences. Data visualization of codes was represented by a word cloud generated in DeDoose.

Authors’ backgrounds included health services research (NL, XZ, AO, HW, and KB), digital health research (NL, XZ, and AO), analytics (XZ), implementation science (NL), clinical psychology (NL, XZ, and AO), pediatric psychology (NL and XZ), bioethics (KB), qualitative research (NL, XZ, and KB), and psychosocial oncology research (NL, AO, HW, and KB).

Cancer was the most commonly mentioned word, as it was included in all hashtags that were used to collect the tweets. Excluding “cancer,” “Bing” lexicon-based sentiment analyses revealed that there were more positive words than negative words in the extracted tweets (Figure 1). The “Bing” lexicon was based on the largest lexicon among our 3 lexicons and was able to analyze the largest number of tweets. The 5 most commonly observed positive words in the “Bing” lexicon and our data set were “support,” “love,” “amazing,” “heaven,” and “happy.” The 5 most commonly observed negative words in the “Bing” lexicon and our data set were “grief,” “risk,” “hard,” “abuse,” and “miss.” Analyses from the “NRC” lexicon showed that most tweets were categorized under the sentiment types of “positive” (N=138,752), “trust” (N=101,036), and “anticipation” (N=100,635). Figure 2A-C displays weekly sentiment scores over a time period representative of the impact of the COVID-19 pandemic for the “Bing,” “AFINN,” and “NRC” lexicons. The sentiment was overall positive. These findings were consistent across lexicons. Based on responses from ChatGPT (version 3.0), the randomly selected subsamples from the pre-, during-, and postpandemic periods demonstrated overall positive sentiment (Textbox 2). Although ChatGPT (version 3.0) analysis was exploratory, findings were consistent with our lexicon-based analyses.

Percent agreement between coders was high (91%) before consensus meetings. Consensus meetings resolved all coding discrepancies. The top 5 codes were awareness (N=449), personal experiences (N=166), research (N=60), caregiver experiences (N=54), patient experiences (N=53), policy and the law (N=52), treatment (N=21), end of life (N=21), pharmaceuticals and drugs (N=17), and survivorship (N=15). Data visualization of code applications in a word cloud using DeDoose software is presented in Multimedia Appendix 2.

Twitter users predominantly used the social media platform to promote public awareness of pediatric cancer. In addition, Twitter users frequently use the social media platform to share personal experiences with cancer. Many accounts were from the firsthand perspectives of caregivers of patients with pediatric cancer in active treatment, bereaved parents, and from patients or survivors of cancer themselves. Twitter was frequently used for health knowledge dissemination of research findings (topics included cancer prevention and risk information, early detection, diagnosis, treatment, and survivorship). Twitter was also frequently used to call attention to and lobby for government programs and federal policies that support pediatric cancer treatment and affordable medical care. Example tweets for the top 10 codes are presented in Textbox 3.

DeDoose software displays the frequency of co-occurring codes using a heat map. Most frequently co-occurring codes are in red, moderately frequent co-occurring codes are in green, and less frequently co-occurring codes are in blue. The code co-occurrence chart of the top 10 codes is presented in Figure 3. Overwhelmingly, awareness and personal experiences were the most frequently cocoded (110 times). Moderately frequent co-occurring codes were caregiver experiences and personal experiences (cocoded 54 times), personal experiences and patient experiences (cocoded 53 times), awareness and patient experiences (cocoded 36 times), and awareness and caregiver experiences (cocoded 34 times).

The purpose of this study was to examine the communication content of pediatric cancer–related tweets and the public sentiment of pediatric cancer tweets. The sentiment of tweets on pediatric cancer was overall positive, revealing the supportive, hopeful, and inspirational messages relayed by patients, caregivers, and other relevant stakeholders. Our findings were consistent with the only other study to examine the public sentiment of pediatric cancer–related tweets [17]. Despite previous research showing predominantly negative sentiments globally during the COVID-19 pandemic [38], our study describes a positive sentiment of pediatric cancer–related tweets during the COVID-19 pandemic. We found that pediatric cancer–related tweets predominantly focused on raising awareness about pediatric cancer and disseminating health knowledge. We found that both patients with pediatric cancer or survivors and their caregivers frequently used Twitter to provide updates on their health status, for social support, and to share messages of hope.

We only identified 2 studies thus far that have examined the pediatric cancer experience on Twitter. The current study expanded on the growing body of literature on social media use in patients with cancer, and early research on pediatric cancer-related use of Twitter. Previous research studies have discussed the importance of including caregiver experiences in addition to those of the patient for individual and family-based well-being and adaptive coping [39-41]. Our findings were consistent with previous studies that have shown that cancer-related tweets center on health communication and social support [36]. Similar to previous studies, there was a diverse array of Twitter users, representing perspectives from patients, family members, oncology providers, and health care, pharmaceutical, nonprofit, and other organizations [2,36,42,43]. Additional studies may use combined sentiment analysis and qualitative approaches to better understand pediatric cancer communication and support resources on Twitter and other popular social media platforms. The current and future studies can help inform the development of novel patient- and caregiver-based social media interventions to improve health knowledge, change health behaviors, and improve health outcomes.

Our study has several limitations. First, our analysis included only tweets in the English language which limits generalizability to populations that do not speak English as a primary language. Second, we analyzed tweets that contained prespecified keywords (ie, hashtags) and may have missed other pediatric cancer–related tweets during the specified study period. Third, we only examined social media use on Twitter which may differ from usage on other popular social media platforms. Fourth, we were unable to identify the account type (organization vs individual, patient or caregiver vs researcher) and extract sociodemographic information of users; this information may have further informed our research findings and the conclusions drawn. Fifth, our qualitative content analysis of Twitter only included a random sample of tweets from the full data set which may not be representative of all pediatric cancer–related tweets during the specified timeframe of our analysis. Sixth, our lexicon-based approaches have inherent limitations. Despite using multiple sentiment lexicons in our analyses, such approaches analyze sentiment based on individual words. We did not expand contractions and apply stemming in our analyses as they were not available in “saotd,” the statistical package we used for data preprocessing. The lack of expanding contractions and applying stemming may have reduced the number of analyzable words and tweets. Twitter users commonly use contractions, abbreviations, slang, and sarcasm. Thus, we conducted supplemental analyses using ChatGPT (version 3.0), a next-generation artificial intelligence optimized for natural language processing, to validate our findings. Although exploratory, these findings were consistent with lexicon-based approaches. Research should further investigate the use of other recent innovative nonlexicon-based approaches that analyze entire sentences, such as embedding-based approaches or transformer-based approaches to analyze tweets related to pediatric cancer. Seventh, our data were global tweets but our specified “pre-,” “during-,” and “post-” pandemic time periods were based on United States lockdowns and timelines. We acknowledge that pandemic timelines differ within the United States and certainly globally. Nonetheless, we think it is important to include all tweets regardless of geographic location for representativeness of experiences due to the pandemic being a global crisis.

Acute, ongoing, and evolving pediatric medical traumatic stress impacts the child in the context of their family, which emphasizes the importance of incorporating the perspectives and experiences of caregivers and other family members [44,45]. Social media use by patients with pediatric cancer, their families, and their medical providers has been well-described [46]. Uses and benefits include opportunities for social support, building collaborative networks, dissemination of health-related information, and treatment recommendations [46]. Researchers have increasingly turned to sentiment and content analyses of Twitter to capture real-time experiences of patients with a range of health conditions and other relevant stakeholders. Such research has included analyses of tweets about various cancer diagnoses.

Twitter, as a popular social media platform, may serve as an effective means for researchers to examine pediatric cancer communication and public sentiment around the world. This study used both quantitative and qualitative methods to examine the pediatric cancer experience on Twitter. Despite the global mental health crisis during the COVID-19 pandemic, we found overall positive sentiment of pediatric cancer–related tweets over a time period representative of the COVID-19 pandemic. The content of pediatric cancer tweets was posted by a range of users and centered on the delivery of health and treatment information, seeking and providing social support, and the amplification of awareness of pediatric cancer. Twitter may serve as a powerful platform for rapid communication with survivors of pediatric cancer and their caregivers, and facilitate the widespread dissemination of patient- and caregiver-targeted behavioral health interventions to improve well-being and quality of life. This would be well-matched to pediatric cancer survivors’ and their caregivers’ current preferences in social media use.