Study Population

Eligible clinicians included oncologists who used smartphones and other devices for internet access, although not necessarily for mHealth. Similarly, eligible patients were those diagnosed with any kind of cancer who owned a smartphone that could access the internet. The survey was administered through qualtrics^XM, a US-based company established in 2002 that allows researchers to conduct surveys in communities that are traditionally hard to reach. To construct the panel, potential respondents were recruited through the Web (using specific keywords) and outlined based on their characteristics, and a stratified sample was later invited to join the research panel.

In particular, patients were selected based on a population panel that provides recruitment via Web (Web banners, pay per click, natural optimization of research, affiliate marketing, email, and online public relations activities). Oncologists were reached out drawing on panels that are constructed by telephone recruitment or via recruitment portals starting from specialized databases—such as those of scientific communities.

The survey was sent to 1800 oncologists and 1800 cancer patients consisting of a random sample of panelists stratified by country and age group. Both clinicians and patients were invited to participate in the study via email through Qualtrics and were provided with a link leading to the survey. The main screen of the online questionnaire provided all respondents with the aim of the study, the investigator information, and the expected time length of the survey (approximately 10 min). The respondents’ right to confidentiality was respected, and consent to participate in the survey was obtained.

One concern when using online recruitment panels is that subjects rush through the online questionnaire without properly reading the provided instructions and questions. To increase the statistical power and reliability of our dataset, we screened respondents based on several criteria. First, we included control questions to detect spammers. Those study participants who failed to answer the control questions, answered all the questions in the same way, or filled in boxes with no-sense comments were excluded from our sample (28 clinicians and 68 patients). Second, we examined the time subjects took to fill out the questionnaire (for clinicians, a mean of 6.31 min and for patients, 6.40 min). Extreme deviations from the average time to complete the questionnaire were treated as outliers and excluded from further analysis. Thus, respondents within the lowest 1% percentile (less than 2.5 min) in terms of total time till survey completion were excluded. Furthermore, we checked whether the subjects’ internet protocol (IP) addresses overlapped. In such cases, duplicate entries from the same IP address were excluded from our analysis (12 respondents—patients—in total).

Variables

The survey instrument included 4 different domains in both the clinician and patient versions: (1) sociodemographic variables (age, sex, education, and salary level), (2) mHealth utilization, (3) mHealth activities performed, and (4) reasons for not using mHealth. Both clinicians and patients were asked about their use of mHealth technologies for the management of cancer. Users, namely, individuals who owned a smartphone or any other mobile device and who used it for cancer-related purposes, were then asked to report for what purposes they used mHealth by choosing from a list of activities. These activities related to different degrees of pervasiveness of the technology aimed at highlighting different user expertise levels, based on a previously designed framework by Nasi et al [10]. As a result, respondents were further classified as either basic (ie, those who used mHealth to schedule appointments, access personal health care information, or read test results only) or advanced users (ie, people who used mHealth to monitor treatment side effects and prevent further events). In contrast, respondents not using mHealth were asked to identify the reasons that had so far hindered them from adopting the technology using 5-point Likert scale items in the following format: 1=I completely disagree, 2=I disagree, 3=I neither disagree nor agree, 4=I agree, and 5=I completely agree.

Statistical Analysis

Descriptive statistics were used to report respondents’ sociodemographic information and the degree of utilization of mHealth in managing cancer care. To measure the relation between specific sociodemographic information and the likelihood of being mHealth users, 2 possible sources of sample selection bias need to be addressed. First, as the survey was administered online, the results are influenced by the general digital divide in the population. Second, the survey was completed only by patients and clinicians using mobile technologies for any purpose, ie, if the respondent could access the internet but was not a user of mobile technologies (eg, smartphones or tablets), the survey was concluded, and no further questions were asked. In the first case, the sample selection bias is relevant but does not influence our results as the target population of mHealth technologies does not include people without basic technological endowments (eg, a computer with internet access). The second source of bias, instead, is more relevant because it refers to the population having access to the Web but whose mobile endowment is low. Ideally, we should not exclude these respondents as they are a part of the potential target of mHealth. In our sample, only 21 out of 2170 respondents reported not using mobile technologies for any given purpose. To account for this potential bias that could still have a potential effect on our results, we used 2 different statistical approaches, namely, a propensity score–based regression adjustment (PSBRA) with weighting and a Heckman probit selection model (HPSM). For both the propensity score equation in the PSBRA and the selection equation in the HPSM, the independent variables were the age group, nationality, and salary level of the respondent. The choice of using 2 different procedures was motivated by the necessity of testing the robustness of the obtained estimations because of the disproportion between censored and uncensored observations. Analyses were conducted with STATA software, version 14 (Stata Corp).

Patient and Clinician Usage of Mobile Health

Of the 2149 participants surveyed, 1153 (53.65%) had previously accessed some sort of mobile technology for cancer-related purposes. Different mHealth access rates were observed in the 2 end-user groups. Among patients, 28.46% (294/1033) were mHealth users: nonusers were the majority in all countries assessed, although there were between-country differences. Clinicians, in contrast, were most often mHealth users: 76.97% of the respondents (859/1116) utilized mobile technology in their daily activity or in the management of cancer patients. The highest percentage was observed in the United States (459/526, 87.26%). Regarding the intensity of use, we observed that among clinician respondents, 32.26% (360/1116) were advanced users and 44.71% (499/1116) were basic users, whereas 18.39% (190/1033) of patients reported being advanced users versus 10.16% (105/1033) of basic users (Table 3).

Clinicians’ and Patients’ Mobile Health Activities

Among the patients classified as mHealth user, approximately half of the respondents used mobile technologies for automation and decision-making support into activities such as scheduling an appointment (157/294, 53.4%), accessing personal information (147/294, 50.0%), and reading test results (135/294, 45.9%). Only approximately one-third of users and, therefore, about one-tenth of total patient respondents, supported treatment and follow-up phases through mHealth by either monitoring side effects (108/294, 36.7% of users), helping prevent further events (85/294, 28.9%), or taking medications as prescribed (97/294, 33.0%; Table 4). With regard to clinicians, the majority accessed mHealth to carry out activities that pertain to the automation and interaction domains: 88.6% (761/859) used mobile apps to perform literature research, 66.9% (575/859) to interact with their colleagues, and 44.6% (383/859) to communicate directly with patients. Fewer clinician users utilized mHealth for decision-making purposes: 46.1% (396/859) used mobile apps to access patients’ electronic health records, 44.0% (378/859) to collect test results, and a smaller number (324/859, 37.7%) of users used mHealth to support decision making for ordering further tests. A minority of users performed activities that support treatment and follow-up in the care process, such as side-effect management (318/859, 37.0%) and compliance monitoring (116/859, 13.5%; Table 5).

Professional Mobile Health Divide Between Clinicians and Patients

Table 6 shows the marginal probabilities of use at means that were 69.8% for clinicians and 38.7% for patients using the PSBRA technique (69.5% and 38.7%, respectively, using HPSM). Other things being equal, clinicians use mHealth more than patients, thus, highlighting an inefficient activation of the complementarities between the two main actors involved in the process of care. Age and salary level influenced mHealth adoption in both end-user groups. Among clinicians, younger professionals exhibited an approximately 15 percentage point higher likelihood of being mHealth users (82.9% vs 64.6% using PSBRA), while this gap was even wider for patients, verging on 30 percentage points. Salary level had a similar impact, with more affluent respondents more likely to be mHealth users than the less well-off respondents. These variables also explain the width of the divide between clinicians and patients. With respect to age, the divide was significantly higher for old respondents (35.4% using PSBRA) than that for young respondents (24.5%), whereas in regard to salary level, the divide was lower for low-income (29.5% using PSBRA) and high-income categories (23.2%) and significantly higher for medium-income ones (between 36.4% and 43.7%). Further differences arose when country-level situations were addressed, with the divide being as high as nearly 50 percentage points in the United Kingdom (51.7% using PSBRA).

Reasons That Hinder Greater Mobile Health Use

Participants who did not belong to the user category were asked about their concerns regarding mHealth use and answered 5-point Likert scale items (Table 7). On the patient side, the most diffused concerns pertained to the preference for traditional means of communication with their doctor (mean 4.26, SD 0.93), the lack of knowledge about the potentials of information technologies (mean 3.82, SD 1.17), and the doubts about the reliability and effectiveness of mHealth for medical purposes (mean 3.03, SD 1.07). For nonuser clinicians, the most substantial doubts were related to the preference for in-person visits (mean 4.13, SD 0.91) and the inability of patients to use smartphones (mean 3.44, SD 0.94).

Incentives for Greater Mobile Health Utilization

To enhance clinician use, several different layers can be approached. The first dimension pertains to artifact design; ideally, cancer apps would need to be designed in such a way that would strengthen the patient-clinician relationship, and they should be tested for validity, accuracy, and self-efficacy to help clinicians and patients orient themselves in what now seems to be an app overload [38].

App designers and developers must do more to bring their end users into the design process. Our findings point to the need for app developers to leverage toolkits to enable patients and physicians to more fully engage in the design and development process, each contributing with their own expertise [39]. This will enable the cocreation of solutions. In working toward the development of a sustainable Information Technology system, it is important to engage the final users, particularly the clinicians and patients, throughout the different phases from problem identification to the design and development phase, aligning the project trajectory to users’ needs and expectations and providing clinicians with the opportunity for self-reflection and revisions. Unfortunately, many mHealth apps are designed without considering the needs of their users in terms of either patients or clinicians [4]. The literature lacks empirically validated guidelines or process models on how to design apps with stakeholders rather than for stakeholders. As a result, a recent overview of systematic studies by Byambasuren et al revealed that most mHealth apps are of low quality [40], which hinders their recommendations by clinicians and their use by patients. In cancer care only, Brouard et al evaluated 117 apps for oncological information and treatment monitoring [41] and found that the validation of those apps was generally poor (27.4%).

First of all, these results suggest that designers and developers need to recognize that a one-size-fits-all approach will not work when it comes to apps dealing with conditions such as cancer. Specific apps that account for differences in types of patients, variance in the stage of the disease, and the kind of care one is receiving, as well as the expertise one has with using mobile technologies, will have better chances for increasing adoption and regular usage.

Second, most health apps lack evidence of clinical effectiveness and do not undergo a formal validation and evaluation process [42]; the lack of evidence on whether and under what conditions mHealth delivers on its promise to improve patients’ health outcomes and the efficiency of the health care process further contributes to restraining greater utilization [29,43-45]. More cancer apps need to be tested for their efficacy as, with few exceptions [31,46,47], the evidence base in support of mHealth technologies is still lacking [9]. However, given that the overall performance of mHealth apps is multidimensional, that is, it can be measured from different perspectives (eg, patients, caregivers, and clinicians), it is necessary to develop a methodological framework to include a wider array of benefits beside clinical outcomes. This is part of the objectives of Pushing the boundaries of Cost and Outcome Analysis of Medical Technologies, a large, 3-year, European Union funded project whose recommendations on how to assess mHealth apps are expected in 2020 [48].

Also, the National Institute for Health and Care Excellence in the United Kingdom has recently started a project aimed at providing guidance on the assessment of mobile technologies (ie, Behaviour change: digital and mobile health interventions) expected to be delivered in 2020 [49].

Third, efficient regulation can help promote the adoption of mHealth apps. mHealth apps are classified as medical devices when they are used for diagnosis, prevention, treatment, monitoring, or alleviation of disease in human beings and for this reason must respond to high regulatory standards for demonstrating clinical benefit and safety [50]. Nevertheless, regulatory systems have rarely been able to catch up with the exponential launch of mHealth apps in the global market and have often been equivocal in establishing whether a software-based technology has to be treated as a medical device. This resulted in a very poor number of clinical trials that included digital health technologies, 860 worldwide in 2017 [51] compared with the number of mHealth apps for managing health conditions in the same period (126,000) [3], which means that the large majority of mHealth apps have entered the market without any clinical evidence in support [52]. This might have reduced clinicians’ and ultimately patients’ confidence in the reliability and efficacy of the apps. The US Food and Drug Administration (FDA) [53] and the Directorate for Health and Consumers of the European Commission [54] have long been trying to clarify the regulatory standards that digital health technologies need to meet and what evidential requirements need to be developed by app manufacturers. Furthermore, the FDA has led a working group within the International Medical Device Regulatory Forum aimed at harmonizing the regulatory framework for software-based technologies across different jurisdictions [55] and will include in the fiscal year 2019 budget a Center of Excellence for Digital Health that will aim at modernizing the regulatory approach to digital health [56]. These efforts are crucial as they would guarantee common rules to manufacturers that work in a global environment and would increase the level of trust in end users.

Fourth, data protection must be addressed to increase end users’ confidence in using mHealth apps. Clinicians still do not feel the full reassurance about the reliability of the data collected and of the available apps [57]. Data should not only be reliable but their usability is also particularly critical: the vast amounts of data potentially available to patients and providers could easily overwhelm them if not put to best use. The FDA has published premarket and postmarket guidance that offers recommendations for the comprehensive management of medical device cybersecurity risks and continuous improvement throughout the total product life cycle as well as incentivized changing marketing and distributed medical devices to reduce risks [58]. More recently, the General Data Protection Regulation, a European Union law, aimed at regulating personal data in the digital world [59]. Although they are too new to be assessed, we think these efforts go in the right direction of increasing clinicians’ and consequently patients’ confidence in using mHealth apps.

Finally, innovative, multidisciplinary home-based models of care are now available for cancer patients who can be actively maintained with oral anticancer drugs and have shown preliminary success in optimally managing adherence during pilot testings [60-62]. Although the impact of personalized mHealth apps on adherence and other significant outcomes of patients on oral anticancer medications is yet to be assessed [63], appropriate economic incentives and related formulas are needed to spur their utilization of these devices [20].

In conclusion, like all other medical devices, mHealth uptake and diffusion largely depend on clinicians’ conviction, but, differently from some other medical devices (eg, implantable devices), the effectiveness of mHealth heavily depends upon patients playing an active role and using it at the same pace as clinicians.

Study Strengths and Limitations

This is the first survey including a large, international sample size comparing 6 different countries in North America and Europe and, even more importantly, covering the two most important end users of mHealth: patients and clinicians. In fact, albeit scant, previous research has primarily addressed cancer patient needs and attitudes toward mobile technologies and not those of clinicians. Although past estimates exist for other specializations [64], to our knowledge, this represents the first evidence of mHealth utilization by clinicians in the cancer field. Second, this is the first study that combines the survey approach with a more qualitative method (workshop and roundtable with key stakeholders) to better interpret and complement the quantitative evidence emerging from the survey to ultimately provide concrete recommendations to decision makers.

However, this study suffers from some limitations that should be considered in subsequent studies. First, the study was based on a volunteer online access panel and, thus, is not entirely representative of the reference population as only individuals who possess some degree of digital competence could be reached and included. However, we believe that the online tool contributed to highlighting the smallest divide between clinicians and patients, which would likely be larger had we not used a digital tool. Moreover, the investigation in the user groups of the activities performed by patients and clinicians who use mHealth was self-reported and not based on actual records of their practice. Finally, this survey presents the limit of generalizability; thus, the divide and the models tested are valid in cancer and cancer supportive care only, and as much as the results are extremely significant, they might not hold true for other types of diseases.

This study was resourced via an unrestricted grant by Helsinn Healthcare. The funders have had no influence on the content of the manuscript. The authors would like to acknowledge Kathi Apostolidis for providing great expertise during the survey draft and pilot and all anonymous respondents who completed the draft survey and helped us improve it with their valuable insights. Further appreciation goes to Francesco De Lorenzo and Alberto Costa for their comments and feedback. The authors also thank Pat Truman and Bets Davis for their valuable contributions to previous versions of the manuscript. The authors state that the manuscript is an honest and accurate account of the study reported here, and no aspects of the study have been omitted.