Implementation barriers for mHealth for non-communicable diseases prevention and management in low and middle income countries: a scoping review and field-based views from

Mobile health (mHealth) has been hailed as a potential Background gamechanger for non-communicable disease (NCD) management, especially in lowand middle-income countries (LMICs). Individual studies illustrate barriers to implementation and scale-up, but an overview of implementation issues for NCD mHealth interventions in LMICs is lacking. This paper explores implementation issues from two perspectives: information in published papers and field-based knowledge by people working in this field. : Through a scoping review publications on mHealth interventions Methods for NCDs in LMICs were identified and assessed with the WHO mHealth Evidence Reporting and Assessment (mERA) tool. A two-stage web-based 1,2 3,4 5


Abstract
: Mobile health (mHealth) has been hailed as a potential Background gamechanger for non-communicable disease (NCD) management, especially in low-and middle-income countries (LMICs). Individual studies illustrate barriers to implementation and scale-up, but an overview of implementation issues for NCD mHealth interventions in LMICs is lacking. This paper explores implementation issues from two perspectives: information in published papers and field-based knowledge by people working in this field.

Strengths and limitations of this study
• The strength of this paper is the combination of explicit and published knowledge with more informal resources through a web-based questionnaire which led to sharing of also negative experiences.
• The scoping review was comprehensive; however, a protocol was not published prior to conducting the review.
• The selection of interviewees for the first round of questionnaires was narrow, limiting the scope and generalizability of findings. The additional second round of an open web-based survey enlarged the scope, although the response rate remained low.
• The field-based views in this paper are not published in main literature but relevant for future implementation of mHealth interventions for NCDs in LMIC

Introduction
Non-communicable diseases (NCDs) such as diabetes, cardiovascular diseases and other chronic conditions are a large burden for societies globally, due to mortality, morbidity and costs [1][2][3] . Effective interventions for prevention and management of major NCD exist, such as screening and early diagnosis, control with essential drugs and self-management support 4,5 . Delivery models for these interventions increasingly include digital channels such as mobile phone applications and websites 6 .
Mobile health (mHealth) has been hailed as a potential gamechanger for NCD management, even in low-and middleincome Countries (LMICs). The technology has great potential to empower patients, health workers and health system managers through applications like self-monitoring devices, electronic information systems and mobile services for follow-up and community support. mHealth tools can improve the performance of health workers through providing online guidelines and referral services 7 . They also help retain patients in care through reminders and self-management through information and measurement tools 8-11 .
The barriers to implementation of mHealth in LMICs are problematic, especially in the domain of NCD management. Many projects remain stuck at the pilot stage and we have limited evidence on effectiveness, cost, and uptake in these settings 12 . Individual studies point to implementation barriers; however, few publications provide a detailed description of the implementation process, implementation barriers and how they were addressed. An overview of implementation issues for NCD mHealth interventions in LMICs is lacking. The mHealth evidence reporting and assessment (mERA) checklist was developed in response to the observed need for increased clarity in reporting and describing mHealth interventions, especially in LMICs. Adhering to reporting guidelines facilitates successful replicability to appropriate contexts.
This paper addresses the implementation knowledge gap, by exploring implementation issues from two perspectives: information in published papers and field-based knowledge on implementation by people working in this field. Our main research questions was: what are the barriers for implementation and scale-up of mHealth interventions for non-communicable diseases prevention and management in low and middle income countries? For this first perspective, we did a scoping review of the literature on evidence in mHealth studies and assessed the reporting on implementation using the mERA checklist. For the second perspective, we collected views from researchers and health workers working in LMIC through a web-based questionnaire. The two methods allow the combination of formal studies and first-hand informal and tacit knowledge. The combination yields a more complete picture of implementation barriers.

Scoping review
The purpose was to determine how many of the studies with the defined key elements report data in line with the current mHealth/e-Health reporting guidelines. The key elements of studies included were mHealth interventions conducted among an adult population in LMICs which had improvement of prevention, detection and management of cardiovascular disease, diabetes, and/or hypertension in LMICs as their primary aim.

Data collection process.
We used the terminology of the WHO initiative on digital health tools for NCDs, the 'Be He@lthy Be Mobile' initiative, to operationalize search terms for NCD prevention and management using mHealth 13 . An iterative approach was used to develop a list of primary search terms, which were: (chronic disease OR non-communicable disease OR cardiovascular disease OR hypertension OR diabetes) AND (mHealth OR mobile health OR mobile). Using the CONSORT-EHEALTH guidelines, the term treatment was interpreted in its broadest sense to include management tools, tools enhancing communication between patient and provider, and tools that provide prognostic or diagnostic information 14 . Our search strategy (extended data 15 ) was guided by the formative methodological framework for scoping reviews 16

Scoping review protocol.
A plan for the scoping review was developed a priori but was not formally translated into a protocol and disseminated.

Web-based questionnaire
The web-based questionnaire was developed to complement information from published literature with field-based knowledge from a broad group of implementers. In a two-step approach, we selected researchers and health workers working in NCD mHealth field. The first phase was data collection among a purposively selected sampling of researchers with hands-on experience with implementation research on this topic and belonging to an international network: the Global Alliance for Chronic Diseases (GACD) 19 . A second phase of data collection was added to widen the sample population to include more researchers and practitioners involved in implementation of mHealth in LMICs, in order to enrich the experiences with other projects. This was done, after the analysis of the first phase of data collection, through an online flash consultation with an open invitation to participate to an African-based Community of Practice of local health system managers and international mHealth experts via the online platform Collectivity and Global Digital Health Network (Web Annex 2) 20,21 . The first phase lasted from Dec 2017-Jan 2018; the second phase from Jan-March 2018.

Data collection tool.
For the first phase, a questionnaire was developed by the research team, and reviewed by two independent researchers for relevance, clarity and completeness. The questionnaire comprised 6 questions on the following topics: 1. Domain of intervention based upon Mecheal et al. 22 (options: improvement of self-management; lifestyle information & health promotion; improvement of quality of care; addressing health system barriers), 2. Implementation barriers and 3. The way they were addressed, 4. Actors engaged, and 5) Priorities to address for scale-up of the project, and 6. General priorities in mHealth and NCD. It was a semi-structured questionnaire with a combination of closed questions (with options to select) and open-ended questions allowing for additional qualitative information. Although the questionnaire was not pre-testing before administration with the target group, the iterative process of approaching the respondents for more clarity allowed for refinement of answers when original questions were not clear. The data were collected by self-completion, followed through email exchange on clarification if information was not clear. For the flash consultation in the second phase, a more general open-ended questionnaire was used (extended data 15 . This was done to keep the invitation to participate open for a broad response. In the subsequent email exchange with the respondents to the flash consultation, specifications were asked to clarify answers. Response rate. Researchers/implementers from 8 out of 13 projects (61.5%) responded in the first round. Six researchers/ implementers responded in the second round in the second (two were excluded because of insufficient data). A response rate was could not be calculated for the second round because of the open invitation to participate did not capture non-respondents. For most projects (8 out of 12), the questionnaire was answered independently by two researchers/implementers working on the project. The majority of the projects (83.4%) were research projects and two projects (16.6 %) were privately initiated by a telecom operator or a health care provider.

Data analysis.
Analysis of the responses was done through a deductive approach led by the first and last author (JVO, KY). The analysis started from pre-identified themes -inspired by mERA and expert guidance from the author. Responses were classified and information was further reviewed for additional themes. Interpretation of the responses was validated through sharing the draft manuscript with respondents. Six out of the 12 respondents contribute to the text with adding detail, examples and clarifications.

Patient and public involvement
The underlying studies have developed their intervention in a participatory way. The dissemination is arranged for each study separately. Patients were not involved in the design of this study, nor in the recruitment and conduct of the study.

Scoping review
In total, 18 studies out of 185 papers were included in the final review ( Figure 1). The studies originated from a variety of countries including: India, Pakistan, China, Tibet, Mexico, Iran, Cambodia, Democratic Republic of Congo, Philippines, Tanzania, Bangladesh, Chile, and Malaysia. More than half of the studies utilized SMS (text messaging) as their main intervention tool. Most of these were patient-centered interventions. One study targeted health providers using a mobile decision support platform. While the majority of the interventions focus on diabetes as the core condition, others addressed heart failure, acute coronary syndrome follow-up, cholesterol risk assessment, foot ulcers, and drug adherence. The majority of studies were relatively small with sample sizes ranging from 48 to 3393 participants. In terms of effectiveness, 15 studies showed improvements in clinical patient related end-points. The few papers that commented on barriers noted the participation to be low, and lowering over time 24,25 . Interventions that combined SMS with personal phone calls or with interactive feedback on patients input reported on patient satisfaction 33, 35 . A lack of information on cost is a barrier to further scale-up.

Questionnaire
Respondents from 12 projects, all of which were implemented in LMICs, answered to the questionnaire. The interventions were comparable, mostly focusing on health education, self-management and quality of care (and or a combination of these aims.) ( Table 2). Seven projects mentioned the intention to address health systems barriers, such as data collection or supply chain management (extended data 15 ). The projects from the first round were research projects in public health care settings (the supply side), whereas two other projects were private entrepreneurial initiatives that focused on the user (demand side). Many projects shared common barriers and also mentioned similar strategies to address them. Effective strategies across the projects were the ongoing engagement with patients and the adaptations in the way of delivering the message. The findings from the survey identified important domains for decision making in the implementation and scale-up of mHealth interventions: 1) reviewing the need for adaptation of the intervention; 2) integrating the mHealth intervention with other digital systems and with the physical health care process; and; 3) designing sustainable scale-up models. Respondents also identified criteria for decision-making and specified them for their NCD intervention: (a) age, literacy, impairments, expectations and financial means of end users; (b) objective of the intervention (education, self-management, quality of care, access to treatment or follow-up); (c) actors involved (patients, caregivers, managers, health workers, pharmacies, etc.); (d) resources and organization of the health system (treatment and support options, patient records, access and quality); and (e) socioeconomic cultural context (behavioral norms, inequalities and gender roles) (Figure 2). We mapped the recommended and available information on mHealth projects onto these fieldbased findings. The call-out balloons represent the mERA items, on which information should be provided according to the international guidelines. The colored balloons show the items on which information was found in the scoping review. Over half of the mERA items were not covered in the papers that were included in the scoping review.
Barriers and how they were addressed NCD disease related barriers included age, complications leading to impaired physical or mental functioning, the natural progression of disease and disease-related perceptions. Several respondents mentioned that the highly prevalence of NCD among elderly and among people of lower socio-economic status, led to the combination of low digital literacy and low health literacy. This meant that uptake and engagement of people required additional training and guidance to allow familiarization and growing awareness on health issues. Two projects explicitly mentioned that proactive involvement and follow-up of caregivers and assisting patients improved utilization. They were encouraged to read messages and take them along in their caregiving. Solutions mentioned in two other projects were to change the transmission mode, from SMS to voice messaging and pictograms.
Disease perceptions shape the expectations of mHealth interventions. Mental health projects report stigma as a barrier for uptake, which requires attention from the start. Although generally, there is an interest in information on NCDs in LMICs, people also perceive many psychological, physical and cultural barriers to change behavior that cannot be changed through the intervention. Low demand and declining engagement with mHealth interventions with a behavior change objective is frequent, even when services are free of charge to patients. Respondents mention an average time of engagement of 9 months. To increase patients' utilization at start, health promotion campaigns addressing stigma are combined with the marketing of an intervention. To maintain engagement over time, some projects ensure personal contact with end-users at regular times, as a form of extrinsic motivation, or to adapt the intervention.
Barriers related to the mobile intervention. Health education given via a mobile device requires adaptation of the content to the delivery channel. Health education on NCDs and the inclusion of motivations techniques is an area in development and still very limited in many health care settings in LMIC. The translation from personal motivational coaching to a mobile application includes an additional challenge in such a context, since the content, the stratification, and timing of information need to be planned ahead in small bits of information with little options for interactivity. Although there are many apps for smartphones, quality varies and they are not always relevant for people in LMICs. The development of the process algorithm and of the motivational messages was typically done in consultation with medical doctors, experts in health behavior, telecom experts, and patients and local health workers. Only one project mentions involving people from the Ministry of Health. It took on average one year, including testing and validation. Patient-oriented interventions reach patients in their routine of daily life, at work or at home, i.e. beyond the usual setting between health worker and patient. They are mixed with a range of other messages and adverts that attract the attention and might clash with the culture or prevailing ideas in other settings. The amplitude and diversity might lead to message fatigue and dilute the effect of messages.
Technical barriers mentioned relate to the network provider and operating system, the hardware and software and the linking of different digital systems. The variety of network providers and telecom operating structure across regions makes the optimal choice for a platform of data-management and providers difficult, especially in the highly commercialized and volatile telecom sector in many LMIC. The gradual linkage of the project with other functions, or the adaptation of delivery mode (message to voice) usually required adaptations to the technical architecture and renegotiations. Only a few projects have managed to link the project software with existing health management information system or to electronic medical records. The case studies show the gradation in complexity, from one to multiple actors being able to use the same system. Some respondents mentioned that the involvement of different care providers and departments and the impossibility of allowing interaction between different information systems.
Phone related problems related to people having multiple phones, people switching providers, and phones being turned off. These restrains end users receiving or reading messages sent by the project. Respondents mention that tracking systems that monitor delivery of messages and automatically resend, are useful. This software is often embedded in telecom software, and sometimes within software available in open source format. Projects can build upon software that is available from other settings or an open source software platform, but they still require the technical competency to adapt to specific project needs and local users, for instance to allow software to run on old-fashioned smartphones and to adjust websites to low bandwidth internet. While a tablet has more options for intervention design, respondents tell us that many users prefer a phone because of the battery life and usability.
Health care services barriers mentioned relate to the lack of resources, access to care and integration of mHealth interventions into health care processes including the resistance of health professionals to change. The lack of technical possibilities to link digital systems with each other leads to interventions set up in a fragmented fashion, increasing the workload for health workers. Two respondents mentioned that a joint assessment of workflow and workload with health workers increased acceptance, and continuing education and feedback maintained their commitment. Other respondents mentioned that though discussions with health authorities on the need for upgrading public health facilities and through partnering with other health care providers, pharmacies and community organizations, their mHealth project served as a means to expose the unmet need for NCD interventions and could thus contribute to a change process to increase resources for NCDs management and access to care.
Contextual barriers mentioned included security, gender differences and social, economic and cultural factors, including limited access to internet of mobile device. One respondent said that informing participants that project phones contained tracking software further prevented theft. In another project, social and gender inequities in access to mobile phones were addressed by encouraging the common use of phones and by sharing of message content, for instance by message delivery at dinner times and encouraging using the speakerphone. This implementation format meant that the format of messages needed adaptation to a common audience, at the cost of personalization. Several respondents mentioned that timing of messages influences the susceptibility of people, so supply-driven interventions need to consider people's daily routines.
Regulatory barriers reported were the lack of clarity on digital health regulations in many countries, which led to continued negotiation in some project. This barrier was more outspoken for the entrepreneurial projects. The stakeholders that were mentioned were the government, medical professional associations and telecom regulators. Medical professional associations in South Africa took the position not to allow mobile consultations without face-to-face contact, which led the project to consider another country. Telecommunications regulators also influenced the delivery options of mHealth interventions, through their policies on number masking, reverse billing and spam filters. Reverse billing (to reduce costs for end-users) needs the operator to allow for special short codes in the system. Sending messages in bulk led to numbers being identified as 'spam' or 'number unknown', leaving users to not see or not to recognize the message as from their provider.

Priorities for scale-up of interventions
When asked for the conditions for further scale-up of their intervention, respondents mentioned the priorities: having a human resource plan, a financing or business plan and addressing knowledge gaps. According to answers, a human resource plan should include capacity for process management (interaction with end users and monitoring), for software maintenance and digital health information management, and for evaluation cycles with feedback from users -to ensure relevance and quality on the longer term. Answers on a financing plan, included different options for financing models that would allow for scale-up: direct payment by end-users, contributions from the governments or other parties such as non-governmental organizations, inclusion in health insurance schemes or linking with other services for which there is a large demand. The respondents from research projects mentioned aiming to keeping cost for end-users low, whereas initiatives started as a business began with generation of demand from people able to pay such a mobile money service. Some research projects reported to seek for social enterprises to partner with, trying to bridge both worlds. A project that targeted primary care providers applied a businessto-business model, in which clinics pay a subscription to utilize the intervention. To attract customers, they linked the mHealth session to education sessions about quality of care, which increased the demand for the mHealth. While mixed models of private and public mHealth initiatives are rapidly evolving throughout the world, the need for regulation and for standards of quality also becomes more important. The third priority area mentioned was that of knowledge gaps that hamper scale up. The limited evidence of behavior change messages for diet and physical activity in LMIC, and of how to address comorbidities were mentioned. Respondents mentioned the potential of sharing hands-on implementation knowledge in user-friendly and open ways, via interactive web platforms such as the Global Digital Health Network.

Discussion
This paper provides an overview of a literature and a field-based view on the challenges to implement and scale up mHealth for NDCs in LMIC. Our findings show that SMS messaging is a dominant mHealth tool for patient-directed of quality improvement interventions in LMICs, and that publications report little on the health system and on context conditions for implementation and scale-up, such as legal regulations and cost little on implementation barriers. The field-views of implementers collected in the web-survey reveals significant barriers and strategies to address them. The main challenges relate to health service organization and cultural context related to mHealth interventions. This information is relevant for decisions on scale-up of mHealth in the domain of NCDs. The combination of the scoping review and the survey information have resulted in a map of decision axis for scale-up of mHealth and an overview of NCD-specific decision-criteria.
The information in this paper echoes implementation challenges from other papers 42-44 but they add an NCDs perspective. The two systematic reviews on impact of mHealth on NCDs show modest and variable results 45,46 ; and the scoping review in this paper points to implementation information being a missing link. The survey findings shed more light on such implementation challenges. These include: age and functional impairments are barriers to utilization; behavior change is a complex process and people might have low expectations of the benefit of an mHealth intervention which lowers the uptake especially in the long term; chronic disease management frequently involves multiple actors who don't share the same information; the variation in network providers complicates universal access to the intervention.
Methodological considerations. The information for this paper was collected through a scoping review and through a twostage web survey among researchers and implementors. This combination allowed explicit and published knowledge to be combined with more informal resources. The scoping review was limited by the facts that many of the studies were published prior to the development of the mERA checklist. Careful reading and classification yielded a lot of information addressing the mERA themes nevertheless. The limitations of the survey entail the selective group of interviewees in the first round, narrowing the scope of projects. The first and last author who evaluated the responses of the survey are part of the GACD network themselves and knew many respondents of the first round personally. This is likely to have increased the response rate. It might also have led to a selection bias (people knowing the authors being more likely to respond), and to a response bias (likeminded answers). We estimate the latter bias to have been outweighed by the fact that respondent in the first step survey have taken time to formulate in-depth and qualitative responses. The interpretation of responses by the authors might have been colored by the authors' own experiences. The widening of the survey via Collectivity (thecollectivity.org) and related platforms widened the scope and yielded new perspective. The 2-step approach exposed the difference between research projects and projects that started from an entrepreneur perspective. The informal resources were not checked by other sources such as formal study reports, but for most projects, the questionnaire was completed 2 times, independently by 2 researchers/implementers. Responses were checked for consistency and complementarity. The informal aspect of the data collection has lowered the barrier for participation and the responses show openness to sharing, also on negative results. The field-based stories add a realistic view on implementation and expectations of mHealth for NCDs in LMIC. Although literacy will increase over time, the societal inequities in education, in health and in access to resources, will remain. The barriers to health education, motivational strategies and communication in LMIC, need additional study. There will be an increasing mix of private and public business models in the ongoing scale-up initiatives. This indicate the need for the building of a knowledge base on the policy, legal, financial and cultural conditions that allow equitable access to these new interventions and benefits for those in most need in LMIC.

Ethics
The studies which are included in the review and in the survey have been subject to approval by ethics committee. The survey presents aggregate data that was limited to describing research results from an array of funded research projects. We do not have any human subjects' data in the study or analyses, and thus we did not seek ethical approval. All participating projects however, received ethical clearance from their respective institutions and other local authorities (e.g. Ministries/Municipalities) to conduct their own studies. All respondents in the survey gave written informed consent to participate in the survey and were informed about how the data will be used in a publication. The data are not used for any other purpose. 14. Eysenbach G, CONSORT-EHEALTH Group: CONSORT-EHEALTH: improving

Data availability
Include the reason for excluding 1019 studies in the flowchart. Figure 2 is difficult to interpret on its own and has not been explained in detail. It is a complex figure to follow.
Kindly check the paper for grammatical errors. A few are listed below for your reference: In the second line of study selection section; word "for" is missing -"6 authors were responsible determining eligibility and inclusion".
In the fourth line of the same section; it should be "was" instead of "are", as the whole paragraph is written in the past tense, except for this sentence. "The types of studies included are randomized controlled trials, crosssectional studies, case control studies, and cohort studies." In first line of page 4, it should be "tested" in place of "testing" -"Although the questionnaire was not pre-testing before administration with the target group," This paper can be accepted for indexing after the authors have reworked the paper based on the suggestions given above.
Reviewer Expertise: General practice, global health.
I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.