The A Measurement Tool to Assess Systematic Reviews checklist [25] was analyzed before conducting this review, with all methodology being established prior to the review being conducted. A university librarian experienced in the field of systematic literature review methodology was consulted. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [26] reporting guideline was adhered to for this review. Rayyan (Qatar Computing Research Institute) [27] software was used for the search.

Textbox 1 shows the questions that were defined.

The Medline, Embase, and Association for Computing Machinery Digital Library databases were searched. The search string was generated and aimed to provide maximum coverage while maintaining manageability. We defined 4 broad themes for our search. Terms within a theme were combined using Boolean operator OR, as seen in Table 1. Themes were then combined using Boolean operator AND.

Each study recruited from the initial search was evaluated to determine whether it should be admitted for analysis. The inclusion and exclusion criteria are shown in Textbox 2.

Screening of article titles and abstracts was performed by 2 authors independently. In situations where eligibility of a study could not be determined based on abstract alone, the full-text article was retrieved. We executed a full-text review of the remaining studies after title and abstract screening to further analyze appropriateness for inclusion. We analyzed all review articles to identify any other appropriate studies. We also reviewed the reference list of included papers.

The initial search and reference list screening identified 721 studies. After title and abstract screening, 660 were excluded, leaving 61 full-text studies to be assessed. Of these, 28 were excluded, leaving 33 studies included for data extraction. The PRISMA summary of the database search is presented in Figure 1.

A total of 33 studies were included. Of the 33 studies, 21 were from North America (14 from the United States and 6 from Canada), 9 were from Europe, 2 were from Asia, and 1 was from South America. Most studies specified the type of mobile device used by participants. Smartphones were used in 22 studies, tablets were used in 9, smartwatches were used in 1, iPod touch (Apple Inc) devices were used in 2, and 3 studies did not specify. Regarding the operating system, 11 studies used iOS (Apple Inc), 5 used Android, 1 used Windows (Microsoft Corp), and 17 did not specify.

Among the included studies, mHealth apps were used within a wide range of surgical subspecialties, including orthopedics (8 studies), general surgery (6 studies), head and neck (4 studies), transplant (3 studies), pediatrics (2 studies), breast (1 study), vascular (1 study), neurosurgery (1 study), and others/multiple (7 studies).

Functionality was divided into 5 clear categories; 26 studies included monitoring of symptoms or wounds, 8 included educational content, 5 provided a communication platform, 5 included physiotherapy and rehabilitation, and 2 enabled medication management. App details are presented in Table 2.

Study characteristics are presented in Table 3. With regards to study design, 5 studies were randomized controlled trials (RCTs), 25 were prospective noncontrolled studies, and 3 were retrospective reviews. Sample sizes ranged from 4 to 494, with a median of 39 patients and a mean of 81 patients. Follow-up ranged from 30 minutes postoperation to 12 months postdischarge. The follow-up period was less than 7 days in 4 studies, between 1 week and 1 month in 15 studies, greater than 1 month in 9 studies, and not declared in 5 studies.

Regarding the method of usability analysis, usage (ie, monitoring of user engagement with the app) was used in 15 studies and was the only usability analysis employed in 4 studies. Interviews were used in 10 studies. Self-created questionnaires were used in 18 studies. Validated questionnaires were used in 9 studies. Of these, 7 used the SUS questionnaire, 1 used the uMARS questionnaire, 1 used the technology acceptance subscale, and 1 used the Computer System Usability Questionnaire (CSUQ).

We have categorized the domains of usability according to the MAUQ. A total of 28 studies covered ease of use and satisfaction, 11 studies covered system information arrangement, and 18 studies covered usefulness.

Average SUS scores ranged from 76 to 95 out of 100, with a median score of 87. The uMARS score was 4.1 out of 5. The CSUQ score was 2 out of 7 (whereby a score of 1 would indicate greatest usability).

There is significant potential for bias in studies evaluating the usability of mHealth apps. Hidden agenda bias and secondary gains bias were common and seemingly underreported in the literature. Of the 33 included studies, 8 officially reported authors’ conflicts of interest, stating that they held shares in the app. Furthermore, several of the study groups were provided with the apps free of charge [28], which has clear implications on the usability domain of satisfaction; users who have paid for an app might be expected to have higher expectations than those who have been given an app for free. Perhaps more worryingly, a number of groups [38] declared no conflict of interest, despite seemingly being founders of their app.

Nonresponse bias is a further concern. Some studies, such as Pecorelli et al [44], had high response rates (96%) to usability analyses. However, others, such as Nilsson et al [43], had much lower rates (57.5% on day 14), and some [51] did not disclose the proportion of responders. Nonresponders to usability analyses are more likely to have reported poor usability. Therefore, studies with high rates of nonresponders are likely to have inflated usability results.

Population bias is a further issue. Younger audiences are likely to be more adept at using mobile technologies. Therefore, studies that include a younger demographic are likely to demonstrate inflated usability results. Additionally, the generalizability of results from studies [44] that included patients that were not used to mobile technologies may be limited and may change in the future, when greater numbers of older patients are used to mobile technologies.

To our knowledge, this is the first comprehensive systematic review to assess usability of mHealth apps in postoperative management. This review identified 33 studies evaluating the usability of mHealth apps in the postoperative period across a broad range of surgical subspecialties, demonstrating the growing interest in this area. Most of the included studies were derived from the United States and Europe, which appear to be hubs of innovation in the field. Unsurprisingly, smartphones were the most commonly used devices. However, we suspect that wearable devices such as smartwatches, which have additional monitoring capabilities such as electrocardiogram monitors, will play an increasingly important role in the future [61].

With respect to study designs, 25 of 33 studies were prospective noncontrolled trials. There were 5 RCTs, but usability was never a primary end point in these studies. We feel RCTs comparing mHealth apps to normal practice (eg, in-person follow-up, telephone follow-up, or no follow-up) would be particularly beneficial in assessing the domains of satisfaction and usefulness. It has also been suggested that mHealth app interventions are associated with a falsely heightened level of user satisfaction due to patients’ affinities for their digital devices [62]. This could be minimized by comparing postoperative mHealth apps to a sham app. However, we also acknowledge that RCTs have previously been described as an impractical evaluation methodology for mHealth apps, due to their prolonged duration from recruitment to results and their high costs [63].

The methodology for assessing usability was generally poor. The majority of analyses used simplistic self-created questionnaires that asked rudimentary questions focusing on the domain of satisfaction (28/33 studies) rather than other domains of usability. Indeed, only 11 of the 33 usability analyses assessed the domain of system information arrangement. We would argue that formal usability analyses should cover all 3 common domains of (1) satisfaction, (2) usefulness, and (3) system arrangement, according to the ISO definition of usability [18]. Validated questionnaires are helpful in assessing these areas reliably. Only 9 of the 33 included studies used validated questionnaires, most of which used the SUS. The SUS is a Likert scale made up of 10 questions. The average SUS score is 68 out of 100, meaning that all 7 studies that used the SUS scored above average in terms of usability. Although the SUS is a quick and cheap means of assessing usability, it was created in 1986, before the first smartphone or the concept of an app was realized. The SUS has not been validated for assessing mHealth apps. In comparison, the MAUQ was recently proposed and validated for use in mHealth apps in a population of English-speaking adults [64]. This is the gold-standard reference for analysis of mHealth app usability. While scores on the MAUQ have previously been shown to correlate with the SUS, this is not a strong correlation (r=0.643), thereby highlighting the inadequacy of studies that have only used the SUS.

A major concern in these studies is the risk of bias. A number of the studies’ authors have a financial interest in the usability of their apps, with high user satisfaction making adoption by hospitals and investors more likely. Furthermore, devices were sometimes provided free of charge, which could influence the feedback from users.

mHealth apps have significant potential during the postoperative period for encouraging earlier discharge, improving patient engagement, and offering a safety net for early identification of complications. Thorough analysis of usability is critical to the adoption of these novel technologies in the postoperative period; those with poor usability will have little impact in health care. According to this review, usability analyses to date have been substandard. They have focused on satisfaction, a narrow dimension of usability, with simplistic self-created questionnaires. Furthermore, there is a significant risk of bias, given the common conflicts of interest among authors of published studies. We hope this review changes future practice, with researchers undertaking more robust assessments of usability by employing validated questionnaires, such as the MAUQ, in blinded RCTs.