Online health communities (OHCs), also known as online support groups, are 1 of the 3 primary channels for health consumers seeking health information on the web in addition to search engines and health professionals [1]. Numerous studies have provided substantial evidence that patients benefit from OHC participation [2-5]. OHCs facilitate information exchange and knowledge acquisition among users. For people with cancer and their caregivers, who have a constant and evolving need for information, OHCs are particularly important for around-the-clock availability, immediate and asynchronous communication, and anonymity [6,7].

Users ask questions on OHCs for knowledge acquisition. Questions posed by patients to acquire knowledge to meet their learning objectives vary in cognitive complexity. The cognitive complexity of learning objectives describes the cognitive skills and abilities the learner desires to achieve. For example, a question seeking advice on treatment decisions from peers (eg, surgery vs biological therapies) is cognitively more complex than one looking for facts in medical directions (eg, how many times a day is a pill to be taken). To identify the cognitive complexity level of learning objectives in OHC users’ questions, this study borrowed the Anderson and Krathwohl taxonomy of learning (A&K taxonomy) [8] from educational psychology. This taxonomy was first proposed by Bloom in 1956 [9] and later revised by Anderson and Krathwohl [8]. As shown in Figure 1, the A&K taxonomy defines 6 levels of learning objectives with increasing cognitive complexity. From low to high (ie, cognitively simple to complex), the 6 levels are remember, understand, apply, analyze, evaluate, and create. The theory assumes that to achieve a higher level of learning objectives, one must master the lower levels.

This study chose 3 levels, understand, analyze, and evaluate, rather than adopting all 6 levels because they are close to real web-based health information–seeking scenarios. As found in the analysis by Cartright et al [10], of queries from web search engines, there are 2 representative web-based health information–seeking intentions: evidence based and hypothesis directed. With the evidence-based intention, one mainly focuses on locating information regarding signs and symptoms, which can be mapped to the understand level of learning. The hypothesis-directed intention, which drives individuals to draw connections and discriminate among different uncertain situations and conditions, aligns with the analyze level. Finally, the evaluate level corresponds to the decision-making intention, which involves seeking information to make a treatment decision.

Reciprocity is another substantial benefit of OHCs [11,12]. Knowledge building and collaborative knowledge production take place through discourse among members of OHCs [13]. Peer users of the community, who usually face the same health condition and endure a similar experience, can provide social support by replying to the initial questions and follow-up discourse [3,14]. We focus on the 2 most frequently exchanged types of social support in OHCs: informational support (ie, offers information, such as the course of the condition, treatment, finance, and insurance) and emotional support (ie, expresses emotions such as caring and concern) [5,6,15].

Because OHCs are a promising learning resource for patients and caregivers, an in-depth study of users’ learning objectives and the corresponding support they receive is needed. First, it must be examined whether patients and caregivers use OHCs to achieve cognitively complex or simple learning objectives. Topics and health conditions discussed in OHCs may affect the patients’ learning objectives. Savolainen [16] found that >70% of the questions in OHCs for depression sought an opinion or evaluation of an issue, resembling a high-level learning objective, whereas contrasting results were found in an OHC for alcoholism, where approximately 50% of the posts looked for factual information that serves low-level learning objectives [17]. However, there is scarce literature regarding the learning objectives of users of OHCs for cancers. To deal with the numerous physical and psychosocial consequences of survival, patients with cancer and their caregivers have been using OHCs to address various cancer-related information needs and gain knowledge about cancer [18-20]. An examination of the learning objectives of people with cancer will add to the empirical knowledge on how OHCs facilitate knowledge acquisition for patients with different health conditions.

Second, it is unclear whether all levels of learning objectives are well supported in OHCs. Higher levels of learning objectives (eg, evaluate) are more difficult to achieve than lower levels of learning objectives (eg, understand) and require support from skilled and knowledgeable peers [17,21]. In this study, we examined the type and amount of support for different levels of learning objectives by measuring the corresponding social support qualitatively and quantitatively.

Third, we are interested in investigating how users’ learning objectives change during their participation in an OHC. Moreover, if one’s learning objective is well supported by peers in the OHC, will this drive them to modify their learning objective to ask a more cognitively complex question in the OHC? The answers to these questions will shed light on the effectiveness of OHCs and the designing of OHCs as web-based learning resources.

Therefore, this paper seeks to answer the following research questions (RQs):

To answer these RQs, we collected 10 years of discussion threads from an OHC for patients with ovarian cancer (OvCa) and caregivers. Because OvCa is a rare cancer [22], health information seeking on the internet can be particularly challenging because of information scarcity and limited public awareness. In addition, OvCa is the deadliest cancer among women [22]. The 5-year relative survival rate of patients with OvCa from 2011 to 2017 in the United States was 49.1% [23]. For individuals with OvCa and their families, managing this cancer can be stressful because of intensive treatments and high rates of disease progression [24]. Owing to limitations in early detection, OvCa is often diagnosed at late stages when the likelihood of cure is low. In the United States, it is the most common cause of death due to gynecological malignancies [25]. People with OvCa use OHCs to address their OvCa-specific, treatment-related, and coping-related information needs [19]. However, owing to a lack of disease awareness, 69% of the patients with OvCa had not heard of or knew nothing about OvCa before their diagnosis, thus making the knowledge acquisition and learning process extremely difficult [26]. Furthermore, studies of people living with OvCa are relatively limited, although people with OvCa need a lot of support. There is a dearth of research investigating what information individuals with OvCa who use OHCs wish to acquire and what support they receive. The findings of this study also contribute to the knowledge on how to better support the OvCa community.

We collected data from CancerConnect, an OHC for patients with OvCa, managed by the National Ovarian Cancer Coalition (NOCC). NOCC is a nonprofit OvCa advocacy organization that has devoted itself to educating and supporting patients with OvCa, survivors, and caregivers since its inception in 1991. The NOCC CancerConnect Community is one of the most active OvCa OHCs [27]. It is a peer-supported OHC with the goal of providing an open-access platform that encourages and enhances interpersonal learning via informational and emotional peer interactions. To this end, NOCC allows registered users to participate and contribute to the community in several ways, such as initiating and replying to posts, searching and reading posts and comments, creating profiles, joining groups, and sending and receiving private messages.

This study was reviewed and approved by the Institutional Research Board of University of Pittsburgh (STUDY20040102). In addition, permission was obtained from NOCC to conduct this study.

Our NOCC data set contained 909 OvCa discussion threads posted between June 2010 and December 2020. Each thread was made of 1 initial post and corresponding comments if any. In total, there were 909 initial posts and 14,816 comments.

Figure 2 illustrates the overall data analysis process. We first performed manual annotations on the 909 initial posts to determine whether there was a question articulated in the post. As a result, 560 posts and their 3998 comments were retained for further analysis. The posts without any questions mainly consisted of sharing personal updates, sharing resources, provoking discussions, and providing inspiration. The posts were then coded in terms of the level of the learning objective and OvCa-related topics. For the 3998 comments on the initial posts, we first performed manual annotation on 500 randomly chosen comments to identify the types of social support. Automatic classification models were then trained and applied to predict different types of social support in the remaining comments.

As mentioned earlier, we borrowed 3 levels from the A&K taxonomy of learning [8] to identify the level of learning objectives in the users’ questions. The descriptions of each level of the learning objective and the deidentified example questions are displayed in Table 1. To achieve higher levels in the A&K taxonomy, one must master the lower levels in the hierarchy. Therefore, the 3 levels of learning objectives were coded mutually exclusively. For example, Figure 3 shows a post with the evaluate level of learning objective, as the poster described her situation and sought decision-related information from peers. The real username and user profile image are removed for privacy.

Two coders (YC and KT) applied the coding framework to 100 sample posts to determine the level of the learning objective that best describes the cognitive complexity of the questions. Substantial agreement was achieved between the 2 coders on the 100 sample posts (percentage agreement=0.79; Cohen κ=0.72), indicating an acceptable level of agreement [28,29]. The 2 coders then met to discuss any disagreements. Throughout the discussion, all disagreements were addressed, and no changes were made to the codebook. A coder annotated the remaining posts by using the codebook.

To better understand OvCa users’ information needs at different levels of learning objectives, the topics of the questions in the initial posts were annotated through content analysis. The coding framework was inductively developed by a nurse practitioner by immersing herself in the posts. A coding framework with 13 topics was established initially.

Using this framework, the 2 coders individually annotated all the posts. Questions in each post included 1 or multiple topics. Later, topics that appeared in <10 posts were further grouped into Others. Consequently, 9 codes were used to classify the topics of information needs in the initial posts (Table 2). An acceptable interrater agreement was obtained between the 2 coders, with an average percentage agreement of 0.94 and Cohen κ coefficient of 0.72, ranging from 0.62 to 0.81 across 9 categories [28,29]. The 2 coders discussed and resolved all disagreements and reached an agreement in all cases.

The 2 most common types of social support exchanged in OHC are informational and emotional support [5,6]. In this study, as the aim was to investigate what information users receive as answers to their questions, the informational support provided in the comment was further classified by using the framework proposed by Chuang and Yang [17]. Chuang and Yang [17] identified five types of informational support:

In addition, emotional support was marked if a comment provided empathy, encouragement, or appreciation [12].

All 6 types of social support, including emotional support and 5 types of informational support, were coded in a binary fashion, and a comment could provide 0, 1, or multiple types of support. If no informational or emotional support could be identified, the comment was coded as “Others.” For example, Figure 4 displays 2 comment examples that replied to posts shown in Figure 3. The first comment was coded as “1” for providing a fact and “0” for all other types of informational and emotional support. The second comment was coded as “1” for providing a fact and an advice and “0” for all others.

The social support types provided in the 3998 comments were identified in 3 steps. First, 2 coders coded 150 sample comments to ensure the reliability of the coding framework. On average, an agreement rate with percentage agreement of 0.94 and Cohen κ of 0.84 were achieved, indicating an almost perfect agreement [28,29]. Second, after addressing all disagreements, a coder coded 350 more comments. As a result, a data set of 500 comments was obtained, in which each comment contained a comment text and corresponding support labels. Third, as it would be impractical to annotate all 3998 comments, the decision was made to build machine learning–based classifiers by using the already annotated comments. In total, 6 machine classifiers were built for each support type. A pretrained Bidirectional Encoder Representations from Transformers (BERT) language model [30] was fine-tuned for each classification task. BERT was used because it obtained good classification accuracy with less data on different downstream text classification tasks, such as sentiment and emotion classification [30]. The data set was split into 3 folds with a 70:10:20 ratio for training, validation, and testing, respectively. The accuracy reported in Table 3 is based on the testing fold. The interrater agreement between the 2 coders and performance of the classification models are presented in Table 3. The code for the model and access to our model are listed on GitHub [31]. Finally, the models were applied to predict the social support types for the remaining comments.

Of 909 initial posts, 560 (61.6%) were associated with learning objectives, as indicated by the questions asked in the posts. The following results were based on the analysis of the 560 initial posts with identified learning objectives and 3642 comments that provided at least one type of support.

This study investigated knowledge acquisition by people with OvCa in an OHC. We borrowed three levels of learning objectives from the A&K taxonomy: understand, analyze, and evaluate. The results revealed (1) the distributions and topics of posts at different learning objective levels, (2) the type and amount of corresponding social support at each level, and (3) the influence of social support on changes in learning objectives. The principal findings, contributions, implications, and limitations of this study are discussed in the following sections.

Our results showed that NOCC was mainly used by patients with OvCa and their caregivers to address information needs with low- to middle-level learning objectives. Of the questions, 82.3% (461/560) were either at the understand or analyze levels of cognitive complexity, in which the user initiates a post to pursue basic facts and concepts or connections and relationships among multiple concepts. Notably, only 17.7% (99/560) of the posts with questions were associated with an evaluate-level learning objective, in which the users asked other OHC members to help them make decisions or judgments based on their specific conditions. These results are partially different from the findings in [16], where >70% of the posted questions in the web-based discussion forums sought an opinion or evaluation of an issue, resembling an evaluate- or analyze-level question, whereas the need for factual and procedural information was less common. These conflicting results could be attributed to the different health conditions studied. In in the study by Savolainen [16], the topic of interest in the threads was depression, whereas in this study, it was OvCa, which is listed as a type of rare cancer by the National Institutes of Health [1]. Therefore, the general public lacks disease awareness and education regarding OvCa, and the information is complex and scattered. This might cause OHC users to seek basic facts and concepts at the understand level more often. In addition, the prevalence of analyze-level questions could be explained by the fact that OvCa is a complex disease. Because the diagnosis, staging, and treatment are complex, patients and caregivers have to learn and sort out which information applies to them and which does not. For example, on average, women with OvCa under treatment need to manage 12 concurrent symptoms [32].

Regarding OvCa-related topics, treatment is the most popular topic of interest among all the information needs, regardless of the level of learning objectives. This finding is in accordance with the results in the study by Madathil et al [19], in which treatment-related information was found to be the most sought-after information by patients (41.3%) compared with OvCa-specific and coping information. Data analyses were conducted at the Ovarian Cancer National Alliance, another OHC for OvCa. We identified 9 different topics by using our fine-grained topic classification framework, and the posts were classified in a nonmutually exclusive manner. Treatment was still found to be the most popular topic. This finding further underlines the high demand for treatment-related information and support among people with OvCa. It is also noteworthy that treatment decision accounted for a large share at the evaluate level despite the concern that an OHC might not be an appropriate resource to ask for treatment-related decisions. Such findings add to the demand for research efforts to assess the quality of treatment-related decisions shared by peers in OHCs.

In addition, we examined the type and amount of informational support in the comments, providing a means to study the quantity and quality of information that OHC users can acquire at different levels of knowledge acquisition. In general, users in the NOCC group received the largest number of comments for understand-level learning objective (N=7.68), followed by evaluate-level (N=7.07) and analyze-level (N=5.63) learning objectives. However, the number of comments itself was not enough to reflect the quality and quantity of social support in the OHC because a comment could provide 0, 1, or multiple types of social support; therefore, we classified the types of social support in the comment, especially informational support.

Descriptive results indicated that, in general, the total amount of social support of all types was the largest for evaluate-level learning, followed by understand-level learning, and it was the least for analyze-level learning. For each type of social support, fact was acquired the most compared with other types of support. This result is consistent with the results in the study by Chuang et al [17], which were based on a manual analysis of an alcoholism OHC. Regarding the effect of the learning objective, the results suggest that more advice, opinions, and emotional support were obtained for questions seeking evaluate-level learning. A possible explanation for this finding is that some subjective knowledge, to a certain extent, was needed to support people with OvCa’s information needs of evaluate-level learning. As justified by the interviewees in the study by Harkin et al [2], practical advice shared by peers in OHCs was welcomed by many interviewees, as such information led them on a “journey to become informed.” It is also notable that although the questions with the analyze-level learning objective were the most frequently posted in the OHC, they received the smallest number of average comments and the least amount of almost all types of social support in the comments. Measures beyond the number of comments and support are required to explore this finding in the future.

Finally, we examined multiple posts from the same user, and the results demonstrated that OvCa users’ learning objectives changed during OHC use. This change was reflected by the transition from the current post’s learning objective to the subsequent post’s learning objective. Most of the users who posted >1 post with a learning objective in the NOCC tended to increase their learning objective (70/216, 32.4%) or remained at the same level of learning objective (90/216, 41.7%), as they continued posting and seeking information in the same forum. Furthermore, for users who increased their learning objective in the next post, a larger amount of support in advice, personal experience, opinion, and emotional support was observed in the current post (Figure 8). In other words, receiving more social support might drive the users to acquire higher-level knowledge in the same OHC. Although the result was not statistically significant, this finding adds to previous studies that have demonstrated the effect of social support on member retention and engagement [5,6,33] and contributes new evidence on the potential effects of social support on collaborative knowledge building and generation in web-based communities [13]. In-depth future research promises to investigate the relationship between receiving social support, especially informational support, and knowledge acquisition in OHCs.

As one of the first studies to investigate users’ knowledge acquisition in the context of OHCs, this study presents several contributions and implications to OHCs and the population of the OvCa community.

First, although there is an extensive body of literature investigating OHCs, and it has been proven that patients and their caregivers would use OHCs to post questions and acquire knowledge [12,15,17], little has been done to differentiate knowledge acquisition with different levels of learning objectives and the associated social support provided by peers in OHCs. Our study contributes empirical evidence and demonstrates that user interactions in OHCs can be described and studied from a knowledge acquisition perspective. Not all information needs regarding the underlying cognitive complexity of the learning objectives are identical. Our study also demonstrated that OHC is a promising resource for users to address information needs with different cognitive complexities and that OHCs can help users to improve knowledge if their information needs are well supported with informational and emotional support from peers.

Correspondingly, OHCs ought to recognize the cognitive complexity of the user’s information needs and the underlying learning objective. Importantly, the quality and quantity of social support from peers are critical for users to address their information needs and seek higher-level knowledge. Enhancing patients’ learning objectives is important because pursuing cognitively more complex learning objectives implies higher patient activation—informed and activated patients who actively engage in health care and decision-making. Higher patient activation is associated with better health-related outcomes [34,35]. Given the result that certain types of support were associated with an increase in learning objectives, algorithms or human moderators in OHCs are expected to match the level of learning objectives in the original post with the appropriate types of social support from active peers.

With their social features, OHCs amplify the benefits of a wealth of information as well as the negative emotions shared by peers. In addition, there are concerns about the quality of the narratives shared by patients in OHCs [36,37]. False information and rumors can cause false expectations [2]. To deal with the downside of OHCs, it is suggested that the content be carefully administered by moderators with professional backgrounds. Attention should be devoted to information-seeking posts with high cognitively complex learning objectives such as pursuing judgments and decisions from peers. In addition, some high-quality learning materials can be developed and disseminated via OHCs, as they have been proven to be an active informal learning platform.

People with OvCa have exhibited constant and dynamic information needs, which changes based on the disease trajectory. Concurrently, their knowledge of the disease evolves gradually over the course of the disease trajectory. Most patients with OvCa have little to no knowledge of OvCa before their diagnosis due to a lack of disease awareness [26]. As the trajectory proceeds, they obtain information and gain knowledge through diverse sources, including OHCs [38]. However, the knowledge acquisition process could be extremely difficult because of the lack of OvCa-related knowledge, poor quality of some information available on the web, and inherent characteristics of OvCa [39]. The high prevalence of questions associated with low- to middle-level learning objectives found in this study further confirmed the public’s lack of awareness of OvCa and the community’s lack of disease knowledge.

By contrast, the findings highlighted the benefits of OHC in supporting the OvCa community. Patients with OvCa and caregivers address their assorted information needs in OHC and exchange information and emotional support in the community. In addition, the results based on the classification of OvCa-related topics provide insights into the information needs of people with OvCa, such as the high demand for treatment-related information and support. As there are multiple treatment options for OvCa, a more personalized search system will be beneficial for providing adjusted and dynamic treatment support. The findings provide implications for future health care providers, practitioners, researchers, and developers to design personalized health information systems that will enhance knowledge acquisition and satisfy the unmet needs of people with OvCa.

In addition to the empirical and practical implications of this study, there are several theoretical and methodological implications. First, this study adopted a mixed methods approach, which allowed us to examine both the quality and quantity of the OvCa community’s knowledge acquisition in OHCs. Second, several coding frameworks originated from this study, such as the coding framework for OvCa-related topics and the coding framework for learning objectives. These frameworks can provide future researchers with an approach to unveil the complicated information requirements of the OvCa community.

Regardless of its strengths, this study has several limitations. First, this study was conducted on the NOCC. Although it is a popular OHC for people with OvCa, the results of this study might be biased toward the site used to collect the data. Second, the measurement of users’ learning objectives in this study was limited by the scope of the A&K taxonomy. Only 3 representative cognitive learning levels were selected. Such a design is based on the rationale explained in the Methods section, but we acknowledge that users’ learning and knowledge evolution was oversimplified. Knowledge acquisition is confined to research settings. Little is known about how much the participants learned via other information sources beyond information seeking and support within the OHC. In the future, a complementary obtrusive approach, such as a questionnaire, would help measure patients’ knowledge acquisition more comprehensively. Third, this study only captures OvCa-related topics based on the information needs of patients and caregivers. Other types of supportive care needs, such as interpersonal or intimacy and daily living needs, were not included in the analysis [40]. Finally, this study did not distinguish patients with OvCa according to their disease trajectory, given the scarce data in the NOCC. However, the literature suggests that the information needs of people with OvCa change with the disease trajectory [41,42]. It would be interesting to investigate whether there is a significant effect of disease trajectory on learning objectives and support in OHC. The answer to this question may help researchers and clinicians design interventions that better support patients with OvCa along their disease trajectory.

This work is one of the first to investigate users’ participation in OHCs from a knowledge acquisition perspective through the analysis of a well-known OHC for OvCa. The results demonstrate that users use OHCs to address information needs with different levels of learning objectives, and simultaneously, they can acquire various types of information and emotional support in the comments from peers. Receiving support drives users to pursue higher levels of learning objectives. These findings contribute to improving OHC designs to support the OvCa community.