The Role of Online Social Support in Patients Undergoing Infertility Treatment – A Comparison of Pregnant and Non-pregnant Members

ABSTRACT

The role of social support in the online setting is explored in this study. For this purpose, the posts of infertility treatment patients participating in an infertility treatment online support group between 2002 and 2016 were retrieved. Members who contributed at least 100 words were divided into two groups according to the treatment outcome they reported (pregnancy). The association between the length of group membership, type of support provided, intensity of interaction, active support seeking, overall sentiment and the amount of sadness, anxiety and anger words and the treatment outcome was examined. The findings suggest that online social, in particular emotional, support acts as a buffer between the stressor and the treatment outcome. The expression of anger and initiating of communication by new members diminish this relationship.

Introduction

Couples seeking infertility treatment engage in online social support groups (OSG) mainly to receive and provide informational and emotional support and strengthen their mental health (Aarts et al., 2012; Wischmann, 2008; Zillien et al., 2011). Informational support is defined as providing information to an individual who is seeking information, while emotional support is defined as the evaluation and acknowledgment of an individual’s feelings or providing comfort and encouragement (Sherbourne & Stewart, 1991; Uchino et al., 1996). Informational support chiefly includes sharing information on different phases of assisted reproductive technologies (ART), primarily IVF (in vitro fertilization), from the initial phase including stimulation of the ovaries using fertility drugs to the final phase encompassing waiting for the fertility treatment outcome (Erčulj et al., 2019; Himmel et al., 2009; Van Selm et al., 2008). OSG participants also share information about medications, menstrual cycle and fertility treatments received. As part of emotional support, they express empathy/sympathy and encourage each other after encountering an unsuccessful treatment outcome (Bobicev et al., 2015; Erčulj et al., 2019; Malik & Coulson, 2010) or congratulate each other upon successful fertility treatment (Erčulj et al., 2019). In this paper, the role of online social support in achieving positive treatment outcome – pregnancy – was examined.

The buffering effect of social support in infertility treatment

Meta-analytic studies show that women with higher anxiety and depression throughout their treatment have a lower pregnancy rate than their less depressed or anxious colleagues (Matthiesen et al., 2011; Purewal et al., 2017, 2018). Since anxiety and depression rise with the duration of infertility treatment (Alhassan et al., 2014; Ardenti et al., 1999) and the number of unsuccessful treatments (Maroufizadeh et al., 2015), it is important to explore mechanisms that reduce the anxious or depressive response. According to the buffering effect theory, social support can play an important part in mediating (also buffering) the effect of a stressor on the response or alleviating a stressful response after it occurs (Cohen & Wills, 1985). The buffering effect is clearly supported by studies of symptomatology, where those including infertility treatment patients are no exception. A longitudinal study on emotional response to IVF treatment, for example, demonstrated that increase in anxiety and depression after 6 months of unsuccessful treatment was lower in women who had strong support from their partner and were satisfied with the perceived social support (Verhaak et al., 2005). Beneficial effects of social support were also observed regarding other psychological outcomes, such as infertility-related stress (Gibson & Myers, 2002) and treatment adherence (Moafi & Dolatian, 2014). The latter is very important because persisting with treatment increases the chance of its success – every IVF cycle brings an increase in cumulative live birth rates (Chambers et al., 2017; McLernon et al., 2016). There is some evidence that not just face-to-face, but also computer-mediated social support positively impacts treatment adherence. In one study, interviewed members of an infertility forum declared that online social support had helped them persist with treatment after facing several treatment failures (Lee, 2017). Social support therefore acts on two levels, psychological – lowering stress, anxiety, and depression, and behavioral – influencing determination and treatment persistence, with both leading to the intended health outcome, pregnancy.

Quality of social support

Not all social support is equally helpful. As argued by the optimal matching model (Cutrona, 1990), social support is helpful if it meets the goals of the support seeker and directly addresses the stressor being experienced. In addition, the greatest reduction of emotional distress occurs when support seekers’ expectations are met (Rains et al., 2020). In the online setting, this is possible in two ways. First, the online support group (OSG) members are similar to each other and experience a similar stressor. Second, online support is easily accessible on a nonstop basis, enables asynchronous exchange, and includes a large number of members available to provide novel information with less potential for role conflict (Rains & Wright, 2016). Sharing information (informational support) is one of the main types of social support found in the infertility OSG, but is closely followed by the provision and receipt of emotional support – expressing empathy/sympathy and encouraging or congratulating members who have a positive treatment outcome (Bobicev et al., 2015; Erčulj et al., 2019; Malik & Coulson, 2010). From these two, emotional support seems more closely related to the treatment outcome, as reported by reviewed research on online social support (Rains & Wright, 2016). That immediate and person-centered emotional support is beneficial for emotional improvement is also argued in the comforting model (Burleson & Goldsmith, 1998; Rafferty et al., 2015). Further, online emotional support has a positive, “pay it forward” effect, on the support recipient. After receiving emotional support, they are more willing to also provide the same to others (Kim et al., 2012). By so doing, they further improve their psychological health since not only receiving, but also providing, emotional support was found to have a beneficial effect on health. Positive reframing and improved quality of life were found in both recipients and providers of emotional support (Rains & Wright, 2016). Given that the relationship between the type of social support and health outcome in infertility treatment patients has yet to be studied, one of the questions this research attempts to provide answer is: (1) Is the type of social support associated with the treatment outcome?

Availability of social support

The quality of social support is closely related to its availability. One study showed that support seekers who are new members are less likely to receive an answer to a new thread from experienced members, which influences the amount of social support they receive (Arguello et al., 2006). The perceived availability and the level of social support received is also associated with the frequency of interactions and time spent online (Rains & Wright, 2016). From what is described above, it may be expected that more engaged and proactive members receive a higher level of social support, which might influence the treatment outcome. Two research questions arise from this assumption: (2) How is actively seeking social support (such as starting new threads) associated with the treatment outcome? (3) How is active engagement (the number of posted words as a proxy for time spent online) in OSG discussions associated with the treatment outcome? In addition, the availability of social support is closely related to the emotion expressed in the post of the support seeker. An expression of anger, for example, inhibits the willingness of other members to react and provide support to an angry support seeker (Arslan, 2009; Dahlen & Martin, 2005). As the relationship between the expression of anger in online discussions and treatment outcome has not yet been studied, this research will try to provide an answer to (4) whether the expression of anger is in any way related to the treatment outcome. The receipt by angry authors of less social support might be reflected in the treatment outcome (pregnancy).

Word use as an indicator of psychological state

The use of specific words provides insight into people’s emotional and cognitive state (Chung & Pennebaker, 2011). Depressed patients, for example, use more words related to sadness (Sonnenschein et al., 2018), while joyful authors use more positive sentiment words (Gill et al., 2008). Interestingly, although infertility treatment is highly stressful (for example, Peterson et al., 2014), the overall sentiment found in the OSG discussions is actually positive (Bobicev et al., 2015, 2014). Positive sentiment was also found in other OSG discussions, such as those of patients after hearing loss (Ali et al., 2013) or breast-cancer patients (Cabling et al., 2018). What is at first glance a contradictory finding can be explained by the high share of encouragement, endorsement and gratitude posts, which in themselves are positive (for example, Bobicev et al., 2014). Still, one’s psychological state is reflected in one’s specific use of words. A longitudinal online study of pregnant women, for instance, revealed that anxiety words were more frequently used by women who had experienced pregnancy loss in the past (Schoch-Ruppen et al., 2018). Further, negative sentiment in their posts predicted earlier delivery to a greater extent than self-reported measures, which were unsuccessful in this regard. This suggests word use is a more accurate “window” revealing the psychological state than self-reported measures, which are strongly influenced by self-response bias. These findings lead to question (5) concerning whether also among infertility treatment patients negative sentiment is associated with the treatment outcome. Building on previous research on the role of anxiety and depression in the success of infertility treatment, one would expect that infertility OSG members achieving pregnancy would use fewer words expressing anxiety and sadness than their less fortunate colleagues. Research question (6) is therefore: Is there an association between words expressing sadness and anxiety and the treatment outcome?

Method

Sample characteristics

Infertility support group discussions on the Slovenian forum med.over.net between 2002 and 2016 were retrieved using the program R. Overall, there were 16,068 threads, 132,374 posts, and 13,190 users with a unique username. There were 3,605 users who posted at least 100 words, and these were included in the further analysis. Pennebaker states that this is the minimum text length needed for linguistic analysis (Pennebaker & Francis, 2001). About 5% of participants matching the criteria of 100 words posted in the OSG only once, but of these about one-quarter reported pregnancy. Therefore, the inclusion of participants with only one post seems appropriate. On average, the participants included in the sample were responsible for 32 posts, however 50% of them made 10 posts or less. To assess the gender structure of the users, 100 users were retrieved by simple random sampling and their gender was manually annotated based on their word use (Slovenian language permits the identity of the speaker to be revealed, especially by the use of particular suffixes). The analysis indicated that most of these OSG participants were female (98%). Since systematic information on users’ characteristics was missing and to obtain at least a sense of who these users are, an online survey was posted on the med.over.net Infertility OSG webpage between June and December 2018. Generalization of the survey findings is limited as it is based on the assumption that the current OSG participants are similar to those in the past and due to self-selection bias. All participants of the online survey were women, with their mean (SD) age being 35 (5) years. The mean age (SD) of mothers in Slovenia in 2016 was 30.4 (5.1) years (Statistical Office of the Republic of Slovenia, 2016). Most participants (86.4%) held a university education or higher. Less than half (41.7%) the participants live in a city. The mean (SD) duration of their relationship with their partner is 9.9 (5.1) years. Half of them had been in the infertility treatment procedure for 21 months (IQR: 12–36 months). Statistical data from the University Medical Center Ljubljana, one of three public centers for infertility treatment in Slovenia, indicate the share of women whose cause of infertility is unknown is 10%, although in the survey sample the share is 30% of such cases.

Operationalization

Discussion topics – type of social support

The content of the OSG conversations was retrieved using topic modeling with the latent Dirichlet allocation method. Ten conversation topics were retrieved and manually (based on the content of each topic) assigned to two online support group types: emotional support (one topic) and informational (nine topics) support. The decision to rely on ten topics was made by examining solutions with different numbers of retrieved topics and a ten-topic solution seemed the most appropriate content-wise – topics were not overly segmented, with each topic having clear content. In the second step, the analysis was repeated on the emotional support topic and three emotional-support themes of encouragement, congratulations, and empathy or sympathy were retrieved. The repetition of LDA with different random starting points resulted in a content-wise stable solution as well as manual examination of a sample of the posts that were assigned to each topic. The procedure and topic description is well documented elsewhere (Erčulj et al., 2019). The share of emotional posts for each user with a unique username was calculated and relied on in the subsequent analysis.

Sentiment analysis

A sentiment lexicon was used to calculate the positive-to-negative sentiment words ratio. The lexicon is based on the English sentiment dictionary of Hu and Liu (2004), includes positive and negative sentiment words in Slovenian, and is freely available online (Kadunc & Robnik-Šikonja, 2017, 2016). The positive-to-negative words ratio as a measure of sentiment has already been used in other studies (O’Connor et al., 2010; Wen et al., 2014).

Following the example of the LIWC sentiment dictionary, negative sentiment words present in the text (n = 4698) were assigned to three negative emotional categories: anxiety, sadness, and anger. Three independent annotators categorized negative sentiment words in one of the above categories or in the “not specified” category. All three annotators had at least a university education (two psychologists and one statistician). Inter-rater agreement as measured by Fleiss’ κ equaled 0.31 (95% CI: 0.30–0.32; p < .001). According to Landis and Koch (1977), this indicates fair agreement among raters. Agreement between each pair of raters ranged from 0.31 to 0.54. Agreement was highest for anger words (κ = 0.54) and lower for sadness (κ = 0.33) and anxiety (κ = 0.23) words. Complete agreement among raters was accomplished for 37% of words, partial agreement for 50%, and no agreement for 13% of words. Words for which at least partial agreement between raters existed were included in the emotional dictionary. Overall, 445 words were included in the emotional category of anxiety (for example: fear, uncertainty, threatening, agony, threat, alarm), 1,234 of anger (for example: aggression, arrogance, nonsense, anger, rage, frustration, manipulative, loud) and 353 of sadness (for example: sorrow, pessimistic, loss, barren, numb, desperate).

User classification by reported treatment outcome

The last three posts per user were used in the analysis since users generally report a pregnancy outcome before moving over to the online pregnancy OSG. If they return to the infertility OSG, they do so to encourage the remaining participants by sharing their story of success. An example of such a post is: “I’ve already written to the Pregnancy OSG and will do the same here. I can’t believe it but, after three unsuccessful years, the pregnancy test was positive. I can’t believe it, but I had a brownish discharge and went to the gynecologist, who confirmed the pregnancy”.

At the start of the analysis, 20% of users (their last three posts) were randomly selected and manually annotated for pregnancy. Overall, 28% of users reported pregnancy. This group of users also included those who later miscarried or had an ectopic pregnancy. Afterward, the text was pre-processed to obtain the features used in the following analysis. Only minor text pre-processing was applied: use of lowercase words, unigrams (single word), removal of stop words, punctuation, numbers, and URLs. Words (features) were transformed into numbers by the bag-of-words presentation while five classifiers were probed: naive Bayes, support vector machine (SVM), k-nearest neighbor (kNN), classification tree, and logistic regression. The classifiers’ performance was evaluated using a 10-fold cross validation. The dataset was divided into ten equally-sized folds. The model was trained on nine folds and tested on the remaining fold. This process was repeated ten times so that each fold was used once as the testing fold (on which the performance was tested). The average performance on all folds was then calculated.

The best-performing classifier was logistic regression, resulting in a classification accuracy of 79%. Amongst the predicted pregnant users, 69% were correctly classified as being pregnant (precision), while only half the users who were annotated as being pregnant were identified as such by the classifier used (recall). To improve the recall, the procedure was repeated on the predicted non-pregnant users. Again, the logistic regression classifier performed best with a 76% classification accuracy, 71% precision, and 56% recall. All users classified as reporting pregnancy by the classifier were examined manually, allowing 100% precision to be achieved. All members classified as reporting pregnancy were indeed pregnant. Overall, such members accounted for 27.2% of all 3,605 members.

Association between several OSG characteristics and pregnancy

A multiple logistic regression was used to test the association between the share of emotional posts, of posts starting a thread, of anger, sadness, and anxiety words amongst the negative sentiment words, the positive-to-negative sentiment words ratio, the length of the OSG membership, the total number of words posted and the pregnancy outcome. The highest variance inflation factor equaled 1.1, meaning there was no multicollinearity between the predictors. The total number of posts as a predictor in the regression model was also considered but excluded due to the high (r = 0.96) and statistically significant (p < .001) correlation with the total number of words posted. Associations with p < .05 were treated as statistically significant. The multiple logistic regression analysis was performed using SPSS, version 24.

Results

Characteristics of the forum participants by treatment outcome and the results of the multiple logistic regression analysis are presented in Table 1.

Table 1. Descriptive statistics (median with 1st and 3rd quartile) by reported treatment outcome and results of multiple logistic regression

	NP	P	OR (95% CI)	P
Length of forum participation (years)	0.3 (0.04; 1.3)	0.5 (0.1; 1.7)	1.05 (1; 1.11)	0.048
Overall number of words	249 (147; 607)	339 (167; 869)	1 (1; 1)	0.084
Emotional support posts (%)	6.3 (0; 19)	9.4 (0; 21.1)	1.01 (1; 1.01)	0.001
Starting threads (%)	11.5 (0; 27.3)	10 (0; 24.4)	0.99 (0.99; 1)	0.003
Positive-to-negative words ratio	1.3 (1; 1.6)	1.2 (1; 1.6)	0.82 (0.74; 0.91)	< 0.001
Sadness amongst negative words (%)	6.7 (0; 11.1)	6.8 (3; 11.1)	1 (0.99; 1.01)	0.988
Anger amongst negative words (%)	6.1 (0; 10.6)	5.3 (1.2; 8.7)	0.97 (0.96; 0.98)	< 0.001
Anxiety amongst negative words (%)	9.7 (4.8; 14.6)	10.7 (6.5; 15.7)	1 (1; 1.01)	0.309

*NP = non-pregnant; P = pregnant; CI = confidence interval; OR = odds ratio

Association between availability of social support and pregnancy

Participants who reported pregnancy were more experienced OSG members than those who never reported pregnancy (p = .048). Half the participants reporting pregnancy had engaged in the OSG discussions for more than 6 months, the top 25% even more than 1.7 years and the bottom 25% less than 1 month. Half of the non-pregnant participants had been online for about 4 months or less, the top 25% for more than 1.3 years, and the bottom 25% for less than 14 days (0.04 years). Being online for less time, they had posted fewer words on the forum (median (Me) = 249; interquartile range (IQR) = 147–607) than the pregnant participants (Me = 339; IQR = 167–869), although the association between the number of words posted and pregnancy is non-significant when controlling for other factors (including length of participation) in the regression model (p = .084).

The odds of reporting pregnancy, when controlling for all factors in the model, decrease with the share of posts that start new threads. Non-pregnant users therefore more actively seek information than pregnant users. Further analysis shows that the difference in the share of opened threads in participants being members of the OSG for up to 3 months (less experienced users) is higher among non-pregnant users (Me (IQR): 6.7 (0–30)) than among pregnant users (Me (IQR): 0 (0–25)). Among the more experienced users, this difference is less pronounced (about a 2-percentage point difference in the medians between the two groups).

Association between the quality of social support and pregnancy

The share of emotional support posts is higher among participants reporting pregnancy in comparison to the non-pregnant participants. The latter have more informational posts. Emotional posts comprise encouragement, offering congratulations and empathy/sympathy. Of the three categories, pregnant and non-pregnant users statistically significantly differed in the share of encouraging posts (p = .032). The median (IQR) share of encouraging posts among emotional support posts in pregnant users is 50% (25%–100%) and among non-pregnant users 43.6% (10.5%–100%). Informational support posts include information about IVF treatment, official procedures, situation appraisal, medical examinations, medications, and alternative treatments. Non-pregnant users have a bigger share of informational support posts than the pregnant users, although the median percentage difference is just 3 percentage points in favor of the non-pregnant users. The median share (IQR) of informational support posts among non-pregnant users is 94% (81%–100%) and for pregnant users 91% (79%–100%).

Association between the psychological state and pregnancy

The overall sentiment as measured by the positive-to-negative words ratio is positive for both groups, but the odds of reporting pregnancy drop as the positive-to-negative ratio rises when controlling for other factors in the model (OR = 0.82; 95% CI: 0.74–0.91).

Among all emotions, users differ in their expressions of anger. The odds of reporting pregnancy drop with a bigger share of angry words among the negative sentiment words (OR = 0.97; 95% CI: 0.96–0.98).

Discussion

This research addressed the role of online social support in patients undergoing infertility treatment by examining the association between characteristics of their online social support communication and treatment outcome (pregnancy). Specific research questions addressed the relationship between the treatment outcome and the type of online social support provided or received, the active seeking of online support, overall sentiment of the discussion and negative emotions expressed in the communication.

Around one-quarter (27.2%) of the more engaged infertility OSG members (posting at least 100 words) in the 2002 to 2016 period reported a positive treatment outcome (pregnancy). The results of multiple logistic regression show the chance of reporting pregnancy is statistically significantly higher among members with longer OSG membership (control variable) and a bigger share of emotional support posts (RQ1), when controlling for other predictors in the model. Half of the non-pregnant members had participated in the infertility OSG for about 3.5 months or less, while the equivalent membership length for the pregnant members is half a year. The latter had a higher share of posts with emotional and a lower share of posts with informational social support. From the (sub)types of emotional support, encouragement was found to differentiate between the pregnant and non-pregnant members the most. The two groups also differed in active seeking of social support, where pregnant members initiated a smaller share of new threads (RQ2). The difference in active support seeking between the two groups of members was particularly expressed in the first 3 months of OSG participation and diminished as time passed. A negative association was also found between the positive-to-negative sentiment words ratio and the treatment outcome, suggesting that the posts of participants reporting pregnancy were less positive than the posts of their non-pregnant colleagues (RQ4). Overall, however, positive sentiment prevailed in the retrieved OSG posts. Among negative emotion words, only the share of words expressing anger was statistically significantly negatively associated with the treatment outcome (RQ5), while the association between the share of anxiety and sadness words among the negative sentiment words and pregnancy outcome was not statistically significant (RQ6). Pregnant members had a smaller share of words expressing anger among the negative sentiment words than their non-pregnant colleagues. The overall number of words, which strongly correlates with the overall number of posts, was not statistically significantly associated with the treatment outcome, indicating that these two groups do not differ in active engagement in the OSG discussions (RQ3).

If the length of OSG membership reflects persistence with treatment, then such persisting is rewarded, resulting in the first step in having a child: pregnancy. The relationship between treatment persistence and pregnancy was also demonstrated in other studies with more rigorous study designs (Chambers et al., 2017; McLernon et al., 2016). The role of social support in persisting with infertility treatment was not explored in this study, yet the results of another study reveal that support received from other members of the online community was crucial for the persistence with treatment after facing several treatment failures (Lee, 2017). Of the two main types of social support found in an OSG setting, the mix of informational and emotional support with higher share of emotional support seems to match the participants’ needs better as higher share of emotional support is positively associated with the treatment outcome. Social support addressing the stressor(s) and expectations of the OSG member should according to optimal matching model (Cutrona, 1990) and sensitive interaction system theory (Barbee & Cunningham, 1995) lead to a bigger reduction of emotional distress. This should positively influence the probability of the positive outcome – pregnancy – and support the buffering effect of the social support on the treatment outcome. The positive relationship between emotional support and health outcome in an online setting was also reported elsewhere (Rains & Wright, 2016). The larger share of encouragement posts and smaller share of initiating messages (starting new threads) among the pregnant OSG members, however, suggests that these members were to a greater extent providers of emotional support to other members than active support seekers. This might be due to the reciprocal relationship between support providers and seekers/receivers since recipients of emotional support are more likely to provide emotional support back to their support providers or other OSG members (Kim et al., 2012). The assumption is in line with other study findings reporting a positive relationship between emotional support and treatment outcome, regardless of whether the person is a support seeker or receiver (Kim et al., 2011; Rains & Wright, 2016). At the same time, the pregnant members’ lower inclination to actively seek support suggests that their stressors might be better addressed in the ongoing communication. Not initiating new threads, especially at the start of the OSG membership, saved them from support withdrawal from other members, which is more commonly experienced by new OSG members who are actively seeking support (Arguello et al., 2006).

Although other study findings suggest that perceived availability and level of the social support received are positively associated with the online engagement (in our study measured as the number of posted words) (Rains & Wright, 2016), the stronger engagement was not associated with the pregnancy outcome in our study. The availability of support is related to the expression of anger (Arslan, 2009; Dahlen & Martin, 2005), and our study findings show that a larger amount of words expressing anger are negatively associated with the treatment outcome.

Regarding overall sentiment, the research findings are consistent with others that reported overall positive expressed sentiment in the OSG (for example, Cabling et al., 2018). Still, positive sentiment is more subdued in OSG members with a positive treatment outcome. The negative sentiment words in this group, however, are not due to prevailing words expressing anxiety, sadness or anger (Table 1), but other negative sentiment words. One explanation for the larger amount of negative sentiment words among pregnant members might be that the number of negative experiences pertained to the treatment procedures (such as the extraction of follicles, possible overstimulation) accumulating with time. As pregnant members are engaged in OSG discussions longer than non-pregnant members, it may be expected that their negative experience with the treatment are more pronounced. The lack of a difference between the two groups in the number of words expressing sadness and anxiety is, however, surprising. It would be expected that non-pregnant members would have more words expressing sadness and anxiety since experiencing depression or anxiety are risk factors for a positive treatment outcome (for example, Purewal et al., 2018). The absence of this relationship might suggest that social support buffers the depressive and anxious psychological response of OSG members and breaks the link between the anxiety, depression, and the treatment outcome. This is a strong assumption worth exploring in research with stronger study designs that would also allow for controlling for other important factors, such as type of infertility treatment, lifestyle, age, cause of the infertility, and BMI.

A major limitation of this study is its weak study design lacking information on other important risk factors, which is due to its unobtrusive nature. It still demonstrates, however, the importance of online social support on the path of infertility treatment patients toward pregnancy. The findings recommend the inclusion of infertility treatment patients in OSG communities and may serve as a “manual” listing common mistakes in OSG communication that should be avoided so as to benefit from such communication the most.