The systematic review and meta-analysis were conducted by following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [12], and our study was registered in PROSPERO (an international database of prospectively registered systematic reviews; registration number: CRD42022322072) on April 2022.

We conducted a systematic search of PubMed, Web Of Science, Embase, CINAHL, China National Knowledge Infrastructure, Wanfang (a traditional Chinese literature database), and SinoMed (a Chinese biomedical database). The publication time was limited to the time of database establishment to April 30, 2022. The publication language was restricted to Chinese and English. We conducted the search by using a combination of keywords and MeSH (Medical Subject Headings) terms. For example, our search strategy for PubMed involved using the following search string: “(((medication errors OR look-alike sound-alike medication errors OR high-alert drug error OR drug use error*) OR (medication adherence OR drug adherence OR medication persistence OR medication compliance OR drug compliance)) OR (adverse drug event* OR ADE OR drug related side effects AND adverse reaction* OR drug side effects OR adverse drug reaction* OR side effects of drugs OR drug toxicity)) AND (mobile application* OR mobile App* OR portable software App*OR smartphone Apps OR portable electronic Apps OR portable electronic application).” More research details are provided in Multimedia Appendix 1.

The inclusion criteria were as follows: (1) age≥18 years, (2) at least one chronic disease, (3) an intervention group that used mmApps and a control group that underwent usual care (ie, without using mmApps), and (4) clinical trials.

The exclusion criteria were as follows: (1) the presence of confounding factors in randomized controlled trials, such as mixed methods (a combination of qualitative and quantitative methods); (2) the experimental procedures are not clear and transparent (eg, the intervention procedures or results are not clearly expressed); and (3) duplicate publications and publications for which we were unable to contact the authors.

We created a Microsoft Excel spreadsheet to extract key information from the included studies, including authors, years of publication, countries or regions, participants’ characteristics, intervention groups, control groups, names of apps, functions of apps, intervention durations, outcomes, and measurement tools.

The studies involved participants (aged ≥18 years) with a chronic disease, including hypertension, diabetes, coronary heart disease, arthritis, and other single chronic diseases. However, studies that involved patients with multiple coexisting chronic diseases were also included.

Owing to different intervention measures, participants were divided into control groups and intervention groups. Control groups underwent usual care, such as keeping a medication diary, attending regular follow-up visits, and attending medication safety lectures. Intervention groups used mmApps in addition to undergoing the usual care provided to control groups. These mmApps included, but were not limited to, smartphone mmApps, iPad (Apple Inc) mmApps, and WeChat (Tencent Holdings Limited) mini-programs.

In terms of medication safety outcomes, we focused on medication adherence, which was calculated by examining medication doses and frequencies in patients with chronic diseases to verify whether they matched physicians’ prescriptions. Another important outcome was the rate of ADE reporting, which depended on whether patients reported the occurrence of ADEs. The third outcome was medication errors. Despite using mmApps, there was still the possibility of medication errors; taking the wrong pill or the wrong dose, giving medications to the wrong patient, and taking medications in the wrong manner or at the wrong time were considered medication errors.

The literature quality assessment was based on 7 criteria from the Cochrane Handbook for Systematic Reviews of Interventions, Version 6.3, which was updated by the Cochrane Collaboration in 2022 [13]. The reviewers made a separate judgment for each item (ie, low risk of bias, high risk of bias, or unclear risk of bias). If a study fully met these criteria, the likelihood of various biases was low, and the quality grade was “A.” If these criteria were partially met, the probability of bias was moderate, and the quality grade was “B.” If these criteria were not met at all, the probability of bias was high, and the quality grade was “C.” Two investigators with evidence-based training were invited to simultaneously evaluate the quality of the included literature, and a third investigator was consulted when disagreements occurred. Articles with an overall quality level of A or B were included, and articles with an overall quality level of C were excluded.

RevMan 5.2 software (Cochrane Collaboration) was used to analyze the data. Weighted mean differences were used to analyze the effect sizes of continuous variables, and relative risk (RR) was used to analyze the effect sizes of dichotomous variables. Further, 95% CIs were used to represent the sizes of combined effects. Additionally, heterogeneity was tested. If I² was <50%, the homogeneity was considered to be good, and a fixed effect model was used for the analysis. If I² was >50%, the heterogeneity was considered to be large, and a random effect model was used for the analysis. Heterogeneity was explained in terms of clinical and methodological heterogeneity, and a subgroup analysis was performed, if necessary.

A sensitivity analysis was performed to assess data stability. The degree of stability was calculated by using a different statistical method and excluding large-sample studies from the analysis, so as to verify whether our results were robust and reliable.

We searched 7 databases and obtained 936 articles. The first step was to exclude duplicate articles via NoteExpress software (AegeanSoftware Corp), which left us with 698 articles. In the second step, studies were excluded by screening titles and abstracts, which left us with 90 articles. In the third step, studies were excluded by carefully screening the full text of articles. Finally, we included 8 articles. The retrieval and selection process is shown in Figure 1.

There were 1355 participants in the 8 studies, which were from 5 countries (China, the United States, France, Canada, and Spain). The publication times ranged from January 1, 2014, to December 31, 2021. The median sample size was 169 (range 61-268), and the following chronic diseases were included in the studies: pulmonary tuberculosis [14], renal cancer or prostate cancer [15], influenza [16], multiple sclerosis [17], oral cancer [18], and multiple chronic diseases [19-21]. The median intervention duration was 6 (range 1-25) months. The intervention groups used mmApps, and the control groups underwent usual care (ie, they did not use mmApps). As for the outcomes, 3 studies reported ADEs [16,17,19], 4 studies measured medication adherence [14,15,18,20], and 4 studies calculated the frequency of medication errors [15,17,20,21]. Further details are shown in Table 1.

In all 8 studies, the intervention groups used mmApps, and each app had its own characteristics. Mira et al [20,21] focused on the impact of mmApps on medication self-management among patients with chronic diseases. They invented an app called “ALICE” in 2014, which could provide prescriptions and medical advice, display medication images, and send multiple reminders. In 2015, the mmApp was improved and optimized by adding QR code scanning functions and renamed as “TUMEDICIN.” The app could scan QR codes on medical products and provide information, including information on the purpose of a medication, daily doses, possible adverse effects, and main precautions.

Wei et al [14] implemented an electronic monitoring intervention based on WeChat (a local social software that has achieved large-scale population coverage) that could be used to monitor patient adherence and outpatient clinic visits, report adverse drug reactions, and remind people to take their medicine. US scholars mainly concentrated on medication safety for patients undergoing cancer treatment. Agboola et al [15] developed Chemotherapy Assistant (CORA) to help patients with renal cancer or prostate cancer report and manage their symptoms by improving their self-efficacy. Greer et al [18] used a smartphone mobile app to execute medication plans, and the app included reminders, a symptom reporting function, and patient education.

Wilson et al [16] and Montastruc et al [19] showed great interest in the reporting of ADEs and independently developed CANVAS (an app for automatically reporting ADEs and evaluating user experience) and VigiBIP (an app for spontaneously reporting pharmacy vigilance and drug safety information), respectively. All of the mmApps reported in this study were developed based on real-time communication technology and intelligent automatic identification technology, which simplifies complex clinical practices and provides convenience for medical staff.

All 8 articles reported adequate random sequence generation and thus had a low risk of bias in this regard. Further, 5 studies reported allocation concealment, so the risk of bias was low in this regard, and 2 studies did not mention concealment and were rated as having a high risk of bias in this regard. As for performance bias, 6 studies were not blinded, so they were rated as having a high risk of bias, and the presence of blinding in the other two studies was unclear. Most of the studies (7/8, 88%) reported results and follow-ups, so the risk of attrition bias was low. With regard to reporting bias, 2 studies had an unclear risk of bias because they did not clearly express participant characteristics, and the other studies had a low risk of bias. In summary, all 8 studies partially met the quality criteria, and their quality grade was “B.” Therefore, they were all included in this study. The specific risk of bias and quality evaluation results are shown in Figure 5.

To explore the stability and the degree of stability of our study’s results, we used a different statistical method to analyze the results. It was found that after changing the statistical method for analyzing different outcome indicators, there were no differences in ADE reporting and medication adherence results, indicating low sensitivity and robust and reliable results, as shown in Table 2.

As for medication error results, our original results (RR 1.54, 95% CI 0.33-7.29; P=.58) changed when we removed large-sample studies (RR 0.69, 95% CI 0.38-1.24; P=.21) [17]. The results did not change substantially, indicating low sensitivity and robust and reliable results.

The results of our study verify our hypothesis—mmApps can effectively improve patients’ ADE reporting willingness (RR 2.59, 95% CI 1.26-5.30; P=.009). This is a brand new result that has not been reported by others. Reporting ADEs and near-miss events was considered an important measure for ensuring medication safety, and information about these events is valuable and can be used to prevent ADEs. The Organization for Economic Co-operation and Development [23] recommends that patients should report ADEs, as ADE reporting does not require enormous financial costs and human resources. Further, ADE reporting is excellent in terms of its value; if done well, ADE reporting can reduce the incidence of harm by 15% and decrease economical burdens, thereby saving millions of dollars each year [24]. mmApps have some merits with regard to increasing ADE reporting willingness. First, they facilitate more communication between physicians and patients [25]. Patients ask physicians for help when they encounter professional terms and confusing problems. Second, patients attach more importance to ADE reporting. Medical staff encourage patients to participate in ADE reporting, and as a result, patients generally realize that their medication safety is considered a top priority. Third, the processing of results is transparent [26]. Patients can use mmApps to obtain feedback about submissions regarding conditions, such as whether a submission is successful and how many people submit the same questions. Additionally, it is convenient for patients to be able to browse through an app to see results as they are processed. We hope to explore more functions that help increase the ADE reporting rate among patients, such as measures for improving patients’ enthusiasm through spiritual encouragement and material reward.

Our study results showed that mmApps can improve medication adherence (RR 1.17, 95% CI 1.04-1.31; P=.007). This may be because mmApps reminded patients to take their medicine regularly, provided educational support, and recorded patient histories To some degree, mmApps also strengthened self-effectiveness among patients with chronic diseases and improved quality of life. Degenerative memory, polypharmacy, and comorbidities are common among patients with chronic diseases. As such, they are at high risk of medication nonadherence, and it is not easy to improve medication adherence in this population. Fortunately, we can convey medical solutions through wireless mobile networks, track medications, and even conduct remote consultations with the help of mmApps. Intelligent mmApps empower patients and facilitate self-management at home and abroad. As an auxiliary to physician intervention, mmApps encourage more patients to participate in medical decision-making, thereby improving their disease control capabilities [7]. Further, mmApps are affordable and convenient technologies that rely on existing mobile networks to remotely monitor patients who are difficult to contact or require strict monitoring. Such apps also have the potential to improve the control of risk factors and health conditions. They especially work well for patients with chronic diseases, such as arterial hypertension [27], diabetes [28], and heart failure [29].

Our meta-analysis showed that mmApps had no significant effect on reducing medication errors (RR 0.41, 95% CI 0.13-1.33; P=.58). This finding is different from those of other studies. For example, Baumann et al [30] validated a mobile app that was an appropriate and feasible tool for reducing simple calculation and handling errors in drug administration. Moreover, Siebert et al [31] studied a mobile app that reduced prehospital medication errors by providing simulated pediatric resuscitation education. As for our different results, we thought of the following reasons. First, it is possible that patients were not able to identify medication error types well, resulting in the capture of only a small sample medication errors. Second, the studied mmApps may have been limited in terms of their functionality (eg, a lack of accurate identification functions). For instance, an app that scans barcodes on medicine bottle labels is only useful if a patient uses it before taking the medication. Additionally, such apps only provide specific patient and medication information, and they cannot be used to determine whether a medicine bottle has been opened. However, such apps can reduce medication errors by 54% to 87% if a medicine bottle barcode is available and if patients use these apps properly [21]. The third reason is that mmApps have not yet formed a comprehensive and timely information dissemination network and do not fully cover all chronic diseases, resulting in unequal medical information transmission and the easy omission of medication error cases [32]. Although our research showed that mmApps cannot reduce the incidence of medication errors directly, they can be used to achieve medication safety by reminding patients to take their medications regularly and providing health information. Therefore, to some extent, mobile apps can reduce medication errors indirectly. So far, there is no consensus on whether mmApps reduce medication errors. As such, more and higher-level studies are needed to verify their effects on medication errors in the future.

There may be some limitations to this study. First, with respect to the quality assessment, the quality of the included articles was low, and the overall strength of the evidence was moderate. Second, most studies (5/8, 63%) lacked randomization and double-blinding. As such, stricter inclusion criteria and more rigorous randomized controlled trial studies should be considered in the future. Third, the intervention durations were inconsistent. The shortest study was only conducted for 1 month, which could have had an impact on the results. We hoped that the duration of intervention in the included studies would be at least 6 months. Finally, we did not perform a cost-benefit analysis of mmApps, which is important for helping patients decide whether to use an mmApp. Therefore, we plan to do more research on mmApps’ economic and time costs.

A total of 8 articles were included in this study. We focused on the effects of mmApps on medication safety, and our results showed that mmApps could increase ADE reporting willingness (P=.009) and significantly improve medication adherence (P=.007) but had little effect on reducing medication errors (P=.58). We analyzed several merits of mmApps, with regard to facilitating the willingness to report ADEs; acquired data on how mmApps improved medication adherence among patients with chronic diseases; and found 3 potential reasons for why our medication error results differed from those of other studies.