Papers were searched following a pre-developed protocol in EBSCO using keywords such as “eHealth” and “Emotio* identification” (for the complete search string, see Supplementary File S1, Table 1). All relevant databases (e.g., The Cochrane Library, Google Scholar, PubMed, PsychInfo, CINAHL, ERIC, ProQuest Dissertations, and EBSCO) were searched to include academic and gray literature. The reference lists of retrieved studies, (systematic) reviews, meta-analyses, and eHealth whitepapers and reports were searched as well. No ethnic or geographic limits were adopted. No client or ethical approval was required.

All studies have been screened for publication year (2000 to date) and adopted language (English or Dutch), and the records were deduplicated. The eligibility criteria include (1) participants: adults seeking support; (2) intervention: eHealth; (3) diagnostic criteria: transdiagnostic concept of difficulty identifying, expressing, and/or regulating emotions (e.g., alexithymia); (4) comparator: either face-to-face care or no comparator; and (5) study design: quantitative or qualitative studies. Quality was assessed using the QualSyst tool.

Data were extracted based on a pre-developed protocol. Study characteristics included the country in which the study was conducted, the intervention(s), the comparator(s), the target population, and the sample size. For each study, the instruments used for data collection, the method of delivery of the intervention, and the results of the measurements were reviewed to analyze the effect of the intervention.

The two authors individually critically appraised the quality of the studies using the standard quality assessment criteria for evaluating primary research documents (QualSyst tool) (24). Supplementary File S4, Tables 4, 5 report two separate tables indicating the quality for the quantitative and qualitative studies.

A total of 151 studies were identified based on the search string, with 66 duplicates excluded. The two reviewers screened 85 titles and abstracts independently. The selection underwent thorough approval by the author team, wherein any disagreements were addressed through constructive discussions, accompanied by well-founded arguments, ultimately leading to a consensus. The inter-assessor reliability was 0.70. Figure 2 shows the full flowchart of this systematic literature review. A total of 25 papers were included in the analysis.

Included studies are internationally dispersed, with most of the studies originating in the USA (e.g., (25–27)) (see Supplementary File S2, Table 2). Some studies were conducted internationally through an online community (e.g., (28–30)).

Study designs include six qualitative studies (e.g., (25, 31)), three surveys (e.g., (32–34)), and one case study (35), but the majority of the studies are RCTs (e.g., (36–38)).

The study populations vary but have a predominant medical focus, such as breast cancer patients (e.g., (25, 26, 39)), traumatic brain injury patients (e.g., (40)), patients with mental health problems (e.g., (28, 41)), and employees (e.g., (42–44)).

All 19 quantitative studies were of good quality overall. Results in Supplementary File S4, Table 4 show that the range is from 0.64 (min) to 1.00 (max), with an average score of 0.85, and show seven studies of good quality (0.70–0.90) (e.g., (26, 36, 37)) and nine studies with excellent quality (>0.90) (e.g., (27, 28, 42)).

The scores for the qualitative studies were calculated similarly (24). The quality of the qualitative studies is low, showing a minimum of 0.35 (e.g., (45, 46)) and a maximum of 0.80 (e.g., (31)), with a score average of 0.55 (see Supplementary File S4, Table 5).

There was significant variability in the modalities used in the included studies. Five studies used a synchronous communication mode (e.g., (26, 35, 37)), meaning that the interaction took place in real time, while the majority of studies used an asynchronous communication mode (e.g., (25, 31, 36)).

In total, 12 interventions were text-based (e.g., (25, 31, 47)), 5 made use of videoconferencing (e.g., (26, 35, 37)), and 8 used a communication mix (e.g., (28, 36, 38)).

The study characteristics show significant variations in the international context, study design, and study population, making it impossible to draw overall conclusions. The majority of the studies originate in the USA (e.g., (25–27)), making it difficult to generalize results to other regions of the world.

Overall, all quantitative studies were of good quality, but the quality of the qualitative studies was low. In contrast to this finding, the ratings of quantitative and qualitative studies by the developers of Qualsyst seem to be relatively similar (24), suggesting that the qualitative studies included in this review are of poor quality (e.g., (45, 46)). Qualitative studies are gaining popularity because they provide an in-depth picture of social health problems and individual experiences. However, the lack of information about the study standards for qualitative studies in the health sector leads to poorly conducted, analyzed, or reported studies (51).

The studies on emotion expression are all considered text-based online support communities (e.g., (31)) and describe how much of the conversation contains explicit emotional expression (e.g., (25)). Interestingly, there seems to be some support for a possible disinhibition effect (48), where anonymity lowers the threshold to share emotional experiences online (47). However, no studies investigate whether an observer on the other end of the computer recognizes emotion expression.

The studies on experiencing emotional support solely look at text-based message boards. While emotions are regularly expressed online, participants experience less emotional support compared with face-to-face interactions (33). This finding might indicate that computer-mediated empathy online is lacking beyond the initial expression of emotions.

The results concerning emotion outcomes after the interaction are quite heterogeneous. Different intervention modalities lead to different results on emotion expression, making it impossible to draw any general conclusions.

The results are somewhat more evident when looking at emotion regulation. Emotion regulation seems enhanced after going through self-paced online interventions (28, 30, 42, 50) and interventions via videoconferencing (e.g., (40)). However, comparisons with face-to-face treatment are still lacking. Only one study compared face-to-face treatment with videoconferencing, which found lower emotion regulation skills after the videoconferencing condition (37).

It is difficult to compare the modalities of the included studies with those of other studies, as the field is rapidly evolving and different taxonomies exist (e.g., (6, 52)). However, keeping in mind the framework on computer-mediated empathy (15), the characteristics of the online modality that are most relevant for the expression and experience of emotions are (1) communication richness, (2) synchronicity of the modality, (3) transmission quality, and (4) communication content. While communication richness and synchronicity are specific to the modality used, transmission quality is specific to the internet connection, and communication content can be altered by user customization (e.g., using emoticons and avatars and adjusting the webcam position). Therefore, a description of the modality is insufficient to establish these characteristics for the described interventions. Moreover, some studies used a communication mix (e.g., (28, 36, 42, 50)), making it even more difficult to determine the influence of the modality used on emotional expression, experience, and support.

In addition, no studies on text-based interventions looked at the effects of using symbols, images, emoticons (17–19), or textual paralanguage cues (18, 19) to convey emotions, limiting the conclusions that can be drawn.

The research discussed in this paper can yield significant findings regarding the correlation between computer-mediated care and emotional consequences. Nevertheless, it is imperative to acknowledge the presence of limitations. Firstly, this systematic literature review focused on the eligibility criteria, including adults seeking help with transdiagnostic emotional identification, expression, and regulation. The narrow focus of this study may limit its relevance to other demographics or emotional concerns. In addition, the study explored a restricted number of modalities, such as self-paced interventions and videoconferencing, so it may limit the variety of computer-mediated care options. Chat-based therapies and virtual reality therapy may create different emotional effects.

Furthermore, the available studies show methodological shortcomings. Few studies with quantitative data allowed for statistical comparison of the data (e.g., (28, 34)), and the included qualitative studies were of low quality (e.g., (45, 46)). Furthermore, few studies compared computer-mediated interaction to face-to-face encounters, while no studies compared different technologies with each other.

Future research should describe the modalities used, situational factors, and end-user alterations in sufficient detail to determine the amount of filtering of verbal and non-verbal signals. In addition, studies that investigate how expressing emotions influence how an observer on the other end of the computer experiences these emotions and gives an emotionally supportive response are needed. Moreover, more in-depth analyses of how different verbal and non-verbal cues influence these steps of computer-mediated empathy are lacking.