The electronic databases of ISI Web of Science, Scopus, and PubMed were searched for articles published between 2000 (because of the explosive growth of internet usage around 2000 [30]) and June 2018. A medical librarian (SCH) developed a Boolean search strategy. Then, a doctoral student (JL) carried out the search strategy, which was revised, if required, by SCH. The keywords searched in the Title or Abstract fields related to (1) complaints (eg, comments OR ratings OR suggestions OR reviews OR feedback) and (2) online (eg, free text OR social media OR e-health OR virtual OR internet OR Facebook OR twitter), which were subject to inquiry (see Multimedia Appendix 1).

Studies were considered eligible if they were (a) related to the collection of primary quantitative data about patient complaints; (b) submitted by patients or third parties on their behalf; (c) comments uploaded to PRWs, organization’s websites, or any other online channel; or (d) conveyed in English to facilitate cross-country comparison.

Studies considered ineligible from this research included those that only referred to (a) physician ratings, (b) satisfaction questionnaires, (c) complaints made to non–health care–related organizations, or (d) qualitative analysis without quantifiable themes.

After removal of duplicates, JL screened the titles and abstracts of all the remaining records for relevance. In the next stage, the full-article text of the retrieved results was independently examined by JL and SCH for inclusion. Discrepancies were adjudicated by a senior researcher (JDM). A total of 17 papers were included in this systematic analysis, as illustrated in Figure 2.

Data were initially extracted by JL and subsequently checked by SCH, JL, and JDM, who together carried out the coding phase. The process consisted of the following phases: (1) descriptive and methodological data included in the study were extracted based on the items listed in the People-Centered and Integrated Health Services [33], (2) the number and categories of patient complaints were extracted, and (3) all themes or categories in selected studies were traversed and classified to match the initial categories. Data was analyzed using a thematic analysis approach adapted from the procedure outlined by Braun et al [37]. JL and SCH co-coded the issues of 2 papers to ensure that the coding framework and themes were commonly understood by the research team. JL and SCH independently coded the issues of the remaining 15 papers. In this paper, we achieved high interrater reliability (kappa=.82). Throughout the coding stage, to ensure consistency in categorizing the issue of complaints, 3 researchers (JL, SCH, and JDM) discussed to eliminate divergence.

During the process of synthesizing identified literature, 35 studies were excluded for various reasons, including inability to distinguish between positive and negative comments [2,39], fuzzy quantity [11,16,41], channels of complaints [40,42], or only the number of high-frequency words of complaints mentioned [15,43]. Ultimately, 17 publications were identified as eligible. A wide range of data sources were represented, including (1) PRWs (10/17, 59%), such as RateMDs and China’s Good Doctor website; (2) government-managed health websites, such as NHS choices (2/17, 12%); (3) social networking sites (4/17, 24%); and (4) national online surveys (1/17, 6%). Of the 4 articles referring to social networking sites, 2 included data captured from tweets, whereas one was from Google+ reviews and one from Facebook.

Through our analysis, we found that pertinent articles have emerged since 2012, with a steady increase observed ever since. Most of the research reported in the analyzed studies focused on PRWs or tweets examining country contexts, such as the United States, the United Kingdom, Germany, China, and Canada. From the data collected, it was identified that 10 articles focused on the United States (59%); 4 on the United Kingdom (24%); and 1 article each on China, Germany, and Canada. Most studies did not screen the departments in which the complainant arose (13/17, 76%), whereas 4 articles (24%) paid attention to complaints received from those dealing with a specific illness or encountering specific medical services. The number of complaints reported (or listed in the thematic analysis) in each study varied widely (average 6543, SD 15,547, range 36-57,028, and median 480).

It was noted that the 17 articles included in our sample had different classification criteria, theme terms, and granularity of complaints. Among them, 6 articles (35%) classified complaints based on published categorization schemes, whereas 9 articles (53%) generated their own coding framework from scratch. The coding framework of 2 articles (12%) came from data source organizations. For example, Zhang et al [9] classified complaints according to the stages of medical consultation, which resulted in a 3-layer classification of stages, including medical consultation, diagnosis and treatment processes, and specific complaint attributions. Emmert et al [44] classified complaints according to the object being complained about and generated a 2-layer taxonomy, referring to both the object and specific complaint attributions. We assessed the quality of the included studies against interrater reliability/performance measures; number of layers; number of codes used, and if there were any definitions or descriptions; and examples provided, as shown in Table 1. The results indicated that the quality of studies varied: 8 did not report any classification measurement results and 4 studies used less than 6 codes, overall, whereas 15 did not provide definitions. A total of 13 studies did not provide examples of each theme/class. Thus, the taxonomy for patients’ online complaints is unstandardized, and it is deemed difficult to identify consistent problems arising in patient care.

All papers specified the coders of the complaints; among them, 71% (n=12) were coded by the authors, 12% (n=2) were coded by data curators or investigators, and 18% (n=3) were coded automatically via machine learning techniques. In addition, 9 articles reported intercoder reliability (6 with the Cohen kappa coefficient, 2 with the Spearman correlation or Cronbach alpha, and 1 unspecified).

In total, across the 17 papers, 326 issue codes were used to code 154,762 complaints. Among them, 36 issue codes incorporating 9602 complaints were not classified into our classification framework for their ambiguous meaning of the category (such as others). When classifying all complaint codes, identified from the literature, into our complaint classification code system, we determined that a single code may include a number of new complaint codes that have been assigned individual type codes. The coding results containing concept explanations and issue numbers are provided in Table 2. Patients’ online complaints were seen to fit into the 4 domains, proposed in the a priori framework: (1) prerequisites, (2) patient-centered processes, (3) care environment, and (4) expected outcomes. From the complaints analyzed, it was concluded that patients have the most dissatisfaction with the patient-centered processes (101,586/204,363, 49.71%), followed by prerequisites (50,563/204,363, 24.74%), and care environment (48,563/204,363, 23.76%), with the least satisfied being expected outcomes (3651/204,363, 1.79%).

The prerequisites domain referred to dissatisfaction with the professional skills and knowledge of the health care provider (48251/204,363, 23.61%) and cooperation between professionals in the medical team (2312/204,363, 1.13%). Among the 4 subcategories, the most common referred to comments about attributes of the patient-centered professional (38,314/204,363, 18.75%), which represents the explicit patient-centered personality traits of professionals.

The domain with the most online complaints, patient-centered processes, was represented by a number of categories, as shown in Table 2. Within this domain, the greatest number of complaints related to a lack of open communication of knowledge, personal expertise, and clinical expertise between the patient and the professional (47,385/204,363, 23.19%). The second category care plan based on patient's individual needs collected 40,722 issues (40,722/204,363, 19.93%). The remaining 3 categories contained a small number of complaints, for example, patient participation as a respected and autonomous individual was represented by 4.00% (8186/204,363) of total issues. Among the 14 subcategories in this domain, information, communication, and education accounted for the majority of complaints, representing 22.80% (46,596/204,363) of total issues reported, whereas no complaints were reported on patient autonomy. The most frequently mentioned subtheme in the care environment domain was availability (28,784/204,363, 14.08%), which represented the timeliness of service and the accessibility of medical staff, facilities, and materials. Common problems mentioned in several articles were lengthy telephone calls made by the physician during consultation and difficulties in patients booking an appointment or seeing a clinician. It is worth noting that issues of therapeutic environment emerged in 12 articles. The expected outcomes domain contained complaints relating to physical comfort and physical care (3009/204,363, 1.47%), and emotional support for alleviation of anxiety issues (642/204,363, 0.31%).

We analyzed online complaints from patients based in the United Kingdom, China, the United States, Germany, and Canada because of these being the focus of the studies identified in the literature. The distribution of the 4 domains of online health care complaints across different countries is displayed in Figure 3.

With regard to care aspects that were complained about most frequently, Canadian (33/64, 52%) and British patients expressed the greatest dissatisfaction with prerequisites (34,828/69,746/, 49.94%), as compared with those of other nationalities. Referring to Canadian patient experiences of dental practice, dentists’ professional competence caused the greatest dissatisfaction (20/64, 31%), whereas the grievances of British patients to prerequisites were largely related to patients’ discontent over the attributes of the patient-centered professionals, which contributed to 42.74% (29,810 /69,746) of all complaint issues reviewed from British patients. Patients based in the United States attached significant importance to patient-centered processes (76,349/97,937, 77.96%), especially relating to information, communication, and education (41,133/97,937, 42.00%). German patients expressed negative comments to systemic problems (14,337/31,095), accounting for 46.11% of all complaints identified from German patients. In particular, poor therapeutic environments led to the most-complained-about topic. Chinese patients’ complaints relating to expected outcomes represented 9.82% (542/5521) of the total complaints made, far exceeding the sample average (3651/204,363, 1.79%).

In addition to the aforementioned observations, it was identified that UK patients showed special features. First, they expressed dissatisfaction with the dimensions of care that patients in other countries did not complain about. Such dimensions included family and friends supported as caregivers, needs are anticipated, and transition and continuity of care. Second, UK patients expressed far-less-than-average complaints about certain dimensions of care than those in other countries, such as financial accessibility (0% vs 1.45%) and information, communication, and education (1.94% vs 22.80%). The proportion of complaints from Chinese patients to experts’ professional competence and physical comfort was significantly higher than that of patients from other countries.

The results of this systematic review show that there is not yet a widely adopted taxonomy framework for classifying patient complaints made online in various settings. This means that studies have defined and classified complaints differently, resulting in limited comparability between available studies and research contexts [24,25,57]. To eliminate this gap, we have developed an a priori framework of PCC, which can incorporate patient complaints online. The NHS Institute for Innovation and Improvement found that PCC frameworks (eg, Picker framework) are broadly appropriate for “what matters most” to patients [58]. We have further validated the scope of the PCC frameworks using data from patient complaints online. Our research is considered beneficial for identifying gaps in the evidence base for patient experience expressed online, which has been identified as domains of PCC.

From the analysis of investigated studies, we have identified that the generic themes of prerequisites, the care environment, and the processes were complained about the most or, put it another way, greatly valued by the majority of patients, regardless of where they came from, and should be deemed priorities for PCC. To be more specific, a practicing clinician should be trained more on these dimensions: attributes of patient-centered professional, information, communication, and education.

Although involvement of family and friends is increasingly viewed as an important component of PCC [1,59-61], patients seldom expressed dissatisfaction with it in our research; apparently far less complained about the low participation of themselves. Compared with technical competence, which constituted a fundamental aspect of health care provision [23], interpersonal attributes of professionals were much more likely to be evaluated by patients; the potential implications of this are twofold: first, consistent with the study by Jia Li et al [26], patients’ needs have different hierarchy, which were firstly stated by Maslow [62], and second, patients have uninformed expectation—“patients are not capable or are reluctant to communicate their expectations” [61], which was validated by Rothenfluh and Schulz [63]. Although there were common concerns across different countries’ patients, variation existed in this study as well. Overall, 5 countries included in our study have different health care systems in terms of health care insurance, drug pricing, physician compensation, etc, and lead to disparities in quality of care, care coordination, and physician education: for example, health care in countries such as the United Kingdom and Canada is publicly financed and the coverage is universal; however, the health coverage of America remains fragmented, with numerous private and public sources, as well as wide gaps in insured rates across the US population [64].

In general, UK patients contributed most of the complaints about the attributes of patient-centered professionals. This finding correlates with that of the NHS Institute for Innovation and Improvement [58]. Besides, British patients had a higher pursuit as they voiced high-level needs such as involvement of family and friends (591 issues), anticipated needs (462 issues), feedback mechanisms to measure patient experience (37 issues), and emotional support (642 issues). British patients were also the only nationality to convey discontent over these aspects of care. Given that the NHS has been collecting data on patients’ experience of care for over 10 years [58] and professionals are occasionally accused of being incompetent in satisfying patients’ needs to be treated as a person, eliminating the gap between knowing and doing is crucial. From another perspective, UK patients are assumed to be more informed about PCC by virtue of a variety of regular national health and social care surveys carried out in the United Kingdom, with frameworks of several surveys adopting Picker’s PCC principles being available [58,65]; however, this hypothesis awaits further confirmation.

Despite nearly none of the British patients in this study expressing discontent over financial accessibility (0/69,746, 0.00%) and physical outcomes (376/69,746, 0.54%), approximately 10% (542/5521, 9.82%; 441/5521, 7.99%) of Chinese patients’ complaints referred to these issues. Although the coverage by publicly financed health insurance in China is near-universal, out-of-pocket spending per capita represented approximately 32% of total health expenditures in 2014 [64]. Aside from this, high registration fees, the formidable markup by ticket touts operating in health care locations [9], fees related to excessive tests and treatments [26,66,67], and insurance reimbursement obstacles (cross-regional medical treatment) [67] constitute Chinese patients’ financial barriers. It is likely that physician-dominated decision-making [68], inadequate communication, and patient distrust of doctors [66,68,69] have led to dissatisfaction with their physical outcomes.

Looking into the aspects of care that deeply concerned German patients, we find it necessary to improve the therapeutic environment of health care provision, in terms of privacy and entertainment, as well as maintaining a continuous healing relationship through telecommunication and house visits [44]. Given that studies on patient experience in Germany have not taken these vulnerable care aspects into account [70-72], these dissatisfaction factors should be tested in future research.

In this study, patients based in the United States conveyed great dissatisfaction for information, communication, and education (41,133/97,937, 42.00%), as well as coordination and integration of care (33,520/97,937, 34.23%). Previous work on patient-centered communication demonstrated a positive correlation between skilled physician communication and patient satisfaction [73-75]. Utilizing patient-centered communication guidelines and codes of conduct, such as physicians’ humility and communication training for physicians and medical students, may bring better patient experience and diminish patient complaints [70,73-76]. As for coordination and integration of care, as necessitated when patients encounter long waiting times in hospitals or disorganized operations, previous studies have focused on hospital-level care coordination strategies associated with better patient experience [77]. Besides, it is envisaged that information technologies can reduce the need to craft laborious, case-by-case strategies for coordinating patient care [30,78,79].

Our findings suggest that it is feasible to identify gaps in evidence bases for patient experiences, which have been identified as domains of PCC. It was observed that differences and commonalities coexist across countries, after applying the proposed taxonomy framework, and we found that there is much leeway for the countries of interest to seek improvement in patient-centeredness.

Patient complaints online can indicate weaknesses in the health care system through the eyes of the patients’ themselves. The proposed PCC framework can be applied to analyze the complaints under a wide PCC range of conditions, treatments, and countries. This review has shown significant heterogeneity of patients’ online complaints in different countries, attributable to the diversity in culture, health care institution, and health literacy. Further work is required to apply the framework, using a plethora of data sources, to compare with other services, organizations, and countries or within the health care service over time, that is, a longitudinal study.

All RQs, proposed in the Introduction of this paper, were answered through conduction of the systematic review. Despite certain studies classifying patient complaints online, none were found to include or adopt a credible taxonomy framework. The proposed PCC framework aligns with what patients currently complain about online. By applying the taxonomy, results show that health professionals’ skills and knowledge, open communication of knowledge, and system issues of PCC constitute the focus of online complaints made by patients. In addition, the differences in patient complaints in a multicountry context are discussed.

As always, there are several limitations to this study. First, it was based on searches in merely 3 databases and focused only on currently available peer-reviewed literature; for this reason, we may have missed information in the gray literature. Second, regarding the small number of included articles, because of artificial screening, and the resulting relatively small sample size, our conclusions, especially the country-specific ones, may not be free of overgeneralizations and missing targets, even by taking into account the “community of common destiny.” Finally, our interpretation of the concepts and scope of the various categories of complaints included in the article may not be fully consistent with the authors of the included papers, especially if the explanations or quotes were not given. Considering the limits to the time, space and researcher resources of this study, it is, nonetheless, a worthy trial that merits further exploration.