READ REVIEWER RESPONSES
The prevalence of type 2 diabetes mellitus in the adult population worldwide has doubled from 4.7% in 1980 to 8.5% in 2014 1 . Low- and middle-income countries (LMICs) are disproportionally affected by diabetes, since diabetes-related complications, such as diabetic foot ulcer (DFU), are more frequent in these contexts 1 , 2 . In the US, 60%–70% of people with diabetes will develop peripheral neuropathy 3 . This is important since one in four patients with peripheral neuropathy will develop a DFU, which will increase the risk of foot amputation significantly 4 .
Thermometry is a tool that can identify early signs of foot inflammation, thus providing early signals to enact management and reduce the incidence of DFU and amputation 5 . Three previous clinical trials 6 – 8 and one systematic review have found that the use of thermometry reduced DFU incidence four- to ten-fold among individuals with diabetes at high-risk of developing a DFU 9 . Additionally, one study found that the addition of counselling to promote self-monitoring of skin temperature to standard care is feasible 10 . However, the benefits of thermometry depend on patient adherence to self-assessment, and foot temperature should be evaluated on at least half of the days to effectively reduce the risk of foot ulceration 7 . Yet, adherence could be challenging, especially in LMIC settings. Therefore, novel approaches to improve self-management thermometry adherence are needed. In this context, interventions using short message service (SMS) for diabetes management have been found to be useful to improve self-efficacy, social support 11 , and clinical diabetes-related outcomes 12 .
Other approaches that could prevent foot ulcers include patient’s foot self-care behaviour, annual foot evaluations, knowledge about diabetic foot in health care workers, and therapeutic footwear 13 . Also, in order to prevent recurrent ulcers, it is important to consider the integration or combination of these approaches 14 .
We propose to evaluate the efficacy of a combination of foot thermometry plus mobile health (mHealth)-delivered reminders, using SMS and voice messaging, in reducing DFU in Peru. Our objective was to compare incidence of DFU in the thermometry plus mHealth reminders intervention arm vs. thermometry-only control arm.
This was a physician- and evaluator-blinded, 18-month, randomized clinical trial with two parallel arms and a 1:1 allocation. Details of the intervention and the study protocol have been published elsewhere 15 . We followed the extension of the CONSORT 2010 statement for reporting pragmatic trials 16 .
Although initially planned to follow participants for 12 months, we decided to extent the follow-up period to 18 months to accrue enough DFU events, as we noticed that the frequency of DFU at six months was lower than we expected. Thus, only the extension of the trial follow-up was changed without affecting randomization or assessment rates. There were no other deviations from the original trial protocol.
Participants were recruited at the outpatient clinics of two third-level public hospitals in Lima, Peru; Hospital Nacional Cayetano Heredia and Hospital Nacional Arzobispo Loayza. In some cases, physicians referred the patient to the study fieldworkers to perform a foot evaluation and in other cases fieldworkers conducted an active search for potential participants in the waiting room of the Endocrinology clinic.
Patients were eligible if they: had a diagnosis of type 2 diabetes mellitus; were between 18 and 80 years of age; were in risk group 2 or 3 using the diabetic foot risk classification system as specified by the International Working Group on the Diabetic Foot ([IWGDF], neuropathy and deformity = category 2, history of ulcer and/or amputation = category 3) 17 – 19 ; had a palpable dorsalis pedis pulse in both feet; had an operating cell phone or a caregiver with an operating cell phone; and had the ability to provide informed consent. Patients were considered not eligible if they had current foot ulcers, active Charcot osteoarthropathy, severe peripheral arterial disease, or foot infection.
Our eligibility criteria used IWGDF categories and included people with diabetes at risk of ulceration group 2 and 3. In so doing, rather than focusing only on those at the highest risk for ulceration (IWGDF group 3) we wanted to pursue a pragmatic approach for the prevention of DFU among people with diabetes, thus including also those participants from the IWGDF group 2 category. All previous studies included mostly participants from IWGDF group 3, and only one clinical trial included group 2 patients.
Development and validation of mHealth messages
The content of the mHealth messages was developed and validated with 19 people with type 2 diabetes mellitus. Messages were tested using short open surveys to evaluate the clarity and appropriateness of the messages. These messages were constructed based on a literature review about the characteristics of health education messages, paired with the advice from a specialist in health communication, taking into consideration the reading level of our population and the use of short messages focused on a single idea. We also asked colleagues with previous experience on the use of SMS and mHealth to review the messages before testing them with patients, and changes were introduced after their revision.
We printed all the messages in a single page which was provided to the participants to read by himself/herself. Afterwards, we evaluated each message using the following six questions: (1) Is the message clear?, (2) Could you tell me how would you explain the content of the message to another person?, (3) Is there any word(s) that is difficult to understand?, (4) Is there something that you do not like about the message?, (5) Is there any suggestion to improve the message?, and (6) Would you prefer to be addressed in a formal way “usted” or an informal way “tu”? (see Extended data 20 ).
At the initiation visit, all participants received education about foot care, i.e. etiology and risk factors for the development of neuropathy and ulcers, as well as recommendations for foot care practices and early signs of ulceration; and instructions for the use of the TempStat™ device (see Extended data). This foot care education was done through three videos that were validated by physicians and patients with type 2 diabetes mellitus. The first two videos lasted 8 and 6 minutes and they were related to foot care, whereas the third video lasted 6 minutes and presented the instructions on the use of the TempStat™ device. The three videos were in Spanish and were showed once at the initiation visit, as detailed elsewhere 15 . The device uses liquid crystal technology to provide a visual image of the temperatures (e.g. yellow image represents a higher temperature than blue image) ( Figure 1 ). Frykberg et al. 21 showed that TempStat™ can detect alarm signs, represented by a yellow color change, and the results positively correlate to temperature findings of infrared thermometer, the gold standard of thermometry devices. Another study found that the device identified 74% of serious foot problems 22 .
Figure 1. TempStat.
A) Normal appearance. B) Alarm sign (yellow spot). Source: Visual Footcare Technologies LLC ©, 2013.
One week after enrollment, the TempStat™ was provided to each participant. Fieldworkers instructed the participants to use the device daily and to contact them by phone or SMS if one of the alarm signs appeared in the pads of the TempStat™: two different colors in the contralateral areas of the feet or a yellow spot in any area for two consecutive days. In these cases, the nurse asked about any lesions in the feet as well as the participant’s activity in the last two weeks and provided recommendations on how to decrease activity until foot temperature normalized. Also, in cases where the alarm sign persisted more than one week, an in-person evaluation was performed to assess the patient for infection and/or a masked injury. Additionally, participants were trained to contact the study nurse in cases of dermal lesion of the foot and they were asked to be evaluated promptly by a nurse who was blind to the intervention. When a DFU was confirmed, the study nurse referred the patients to follow the standard protocol.
In the intervention arm, in addition to the TempStat™, participants received the mHealth component weekly (two reminder messages and six foot-care promotion messages each week) for the 18-month study period via both SMS and voice messaging.
Developed and validated messages 23 were sent at 8am approximately and, for the first two weeks of the intervention, daily (Monday to Friday) reminders to use the TempStat™ were sent. Thereafter, for the remaining 76 weeks, patients received only two messages per week at the same time: the content alternating between reminders to use the TempStat™ and promotion of foot care (one SMS and one voice message). Messages were delivered to the participant or caregiver’s cell phones through an automated software system developed by the study team (see Software availability 24 ). Every week the system was evaluated by the study coordinator to verify its functionality.
At baseline, enrolled participants provided information to the fieldworker through questionnaires on lifestyle, history of cardiovascular disease and diabetes, current diabetes treatment, use of insoles, use of orthopedic shoes and mobile phone literacy and underwent a demographic evaluation (age, gender, educational level), socioeconomic evaluation (working status), depression assessment (Patient Health Questionnaire-9), anthropometric evaluation (weight, height and body mass index) and blood pressure measurements (see Extended data 25 ).
Periodic assessments of the participants involving a general checkup and lower extremity evaluation was conducted every two months by the nurse evaluator. Additionally, the nurse collected data about diabetes treatment, caregiver presence, use of insoles and/or orthopedic shoes, and had their weight and blood pressure measured (Extended data 25 ). In some cases, participants could not attend to the hospital for the checkup; in those cases, we completed the visit by phone or by domiciliary visits. In the last visit at 18 months, participants were asked to return their logbook of temperature measurements. In general, participants were encouraged to maintain regular visits with their treating physician in the outpatient clinic.
Glycated hemoglobin (HbA1c) was measured at baseline, six, 12 and 18 months. Measurements at baseline and 18 months were used for the study and measurements at six and 12 months were for standard of care. HbA1c was measured using high-performance liquid chromatography (D10, BioRad, Munich, Germany). The blood sample was collected in the endocrinology clinic by the nurse evaluator during the periodic assessment at the time periods specify above. All samples were transported to be analyzed in a single facility and were checked with regular external standards and internal duplicate assays and monitored by BioRad for quality control.
The primary outcome was DFU. The definition was based on the American Diabetes Association criteria 26 , 27 and for this study it was considered as the presence of DFU occurring at any point during the 18-month study period after randomization. The evaluator was a trained nurse blind to the intervention allocation. The identification of a DFU was through three ways: during the bimonthly clinical nurse evaluations; if an alarm sign had been noted and prompted the participant to seek clinical evaluation; or if the participant identifies a dermal lesion and seeks clinical evaluation.
The following were pre-defined as secondary outcomes: adherence to daily temperature measurement, defined as the participants having recorded their temperature measurements in the logbook on ≥80% of days, and ≥1% reduction in HbA1c when comparing the 18-month with baseline values. Another outcome was alarm signs registered in the logbook.
Our protocol 15 considered one additional pre-defined secondary outcomes: frequency of alarm signs reported to the study nurse. This was not analyzed because of their low frequency. The dose-response analysis of SMS and voice messaging, pre-specified as a secondary outcome in the protocol, was included as part of the process evaluation.
Our a priori sub-group analyses were i) previous foot ulceration and ii) caregiving status, considering assistance provided to the patient with basic activities of daily living, or in the identification, prevention, or treatment of diabetes or any disability. Also, within the intervention-arm only, the type of recipient of the messaging (patient vs. caregivers) was considered for sub-group analyses. In our protocol 15 , we also considered sub-group analyses of participants that use insoles and/or orthopedic shoes, but these were not analyzed due to low frequency.
The sample size was estimated using data from previous randomized trials in study populations similar to our study population 7 , 8 . We expected an absolute change of 21% between the intervention arm and the control arm (9% vs 30%) and with a power of 0.9 and an alpha of 0.05, we required a sample size of 78 participants. We planned to enroll 86 participants in each study arm, anticipating a 10% dropout rate.
We conducted stratification using the hospital site as a single stratum and blocks of 6 to generate a random allocation sequence. Sealed envelopes with codes to randomize participants were used. An independent researcher prepared the envelopes, and the study nurses assigned the codes to each of the enrolled participants. Separately, the study coordinator was responsible for opening the envelopes and informing participants about their intervention or control allocation as per the random list. The nurse/independent evaluators were not aware of the patient's group allocation.
The participants were instructed not to discuss their treatment assignment with the blinded evaluator. Physicians providing care to study participants, nurses and the field coordinators were blind to treatment allocation.
Additionally, we performed a process evaluation during the 18-month follow-up visit to a random group of participants of the two study sites. We obtained information through a set of questions and direct observation of the use of the TempStat™ with 102 participants. In addition, with 39 participants, we asked close and open questions about the messages received in the week prior to the 18-month follow-up visit. As part of this process evaluation, we aimed to know: i) if participants knew how to use the TempStat™; ii) how many SMS and voice messages were delivered by the automated system to study participants according to the automated system; iii) how many SMS and voice messages were received by study participants according to the automated system; iv) if participants understood the messages (only if participants reported that they had received a message in the previous two weeks); and v) opinions from the participants about their preferences in SMS vs. voice messages.
The process evaluation was performed by two fieldworkers different to those who delivered the intervention and data collection was conducted through observation (participants were asked to show how they used the TempStat™), questionnaire (about nursing consultation, report of communication with study nurses, reasons for communication, alarm sign detection) and open questions (related to SMS or voice messaging preferences, use of TempStat™, suggestions about how to improve the intervention) 25 .
To compare the rates of DFU between study arms we performed a time-to-event approximation using Cox’s regression, having time to DFU at 18 months as an outcome. Hazard ratios (HR) and their respective 95% confidence intervals (95% CI) were estimated for the primary outcome of DFU and for the a priori defined sub-group analyses. These analyses included all retained participants, regardless of the number of visits attended, following the intention-to-treat principle. The model was adjusted by site and history of previous ulcer. Evaluation of secondary outcomes of interest was performed using logistic regression analysis to calculate odds ratios (OR) and 95% CI. Data analysis was conducted in STATA V.14.0 (StataCorp, College Station, TX, USA).
For the process evaluation, frequencies and percentages are presented. Also, open-ended questions were transcribed, and then a codebook was created, themes were derived from the data. Coding was performed manually and patterns of answers are described.
The study protocol, informed consent templates, and questionnaires were reviewed and approved by the Institutional Review Board (IRB) at Universidad Peruana Cayetano Heredia (UPCH) in Lima, Peru (SIDISI 61482). In addition, participating hospitals (Hospital Cayetano Heredia and Hospital Nacional Arzobispo Loayza) in the study received the protocol and consent form for approval 16 . The extension in the follow-up period was also approved by the IRB at UPCH and the participants re-consented. The fieldworker explained the study procedures, then the potential participant read the informed consent form and asked questions. After that, if they accepted, they signed the informed consent form. The trial was registered at ClinicalTrials.gov with the identifier NCT02373592 (27/02/2015).
The recruitment was conducted between October 2015 and March 2016 and the follow-up period lasted until October 2017.
In total, 416 participants were screened and 214 were eligible for the study. Of these, 192 gave informed consent and 172 attended the initiation visit and were allocated to the control (n=86) or intervention (n=86) arms ( Figure 2 ). Only 79/86 (91.9%) participants in each arm completed the 18-month follow-up. Reasons for lost to follow-up included migration back to the participant’s place of origin, wrong/incomplete addresses provided, or the participant did not answer the contact phone calls.
Figure 2. Flowchart.
The baseline characteristics were similar between the intervention and control arms, with few exceptions ( Table 1 ). History of previous foot ulcers was reported with more frequency in the intervention arm; 65.9% vs. 48.2% in the control arm (p-value 0.02). Mean HbA1c was 8.9% in the intervention arm and 8.2% among the controls (p-value 0.03). In terms of mHealth literacy, there were no major differences between study arms, with the exception that participants in the intervention arm reported more frequently never having problems with cellphone coverage (89.5% vs. 74.4% in the control arm, p-value 0.01).
Table 1. Baseline characteristics.
Avoid utilization of tight shoes
Sub-group analysis within the intervention arm
Participants were arranged according to the recipient of the mHealth reminders; the participants themselves (45/86) or the caregiver (41/86). We found no evidence of a difference in DFU incidence between these two groups in crude (HR 1.09, 95% CI 0.44 – 2.70), and adjusted analyses (aHR 1.72, 95% CI 0.65 – 4.54, adjusted by site and previous ulcer).
Process evaluation indicators
Some process evaluation indicators for TempStat™ use and understanding of the messages are shown in Table 4 and Table 5 . This data was obtained at the 18-month follow-up visit 29 , 30 .
Dose of the mHealth component. The total number of messages to be sent to the patients in the intervention group during the study period was intended to be 86 text messages and 76 voice messages. The automated software system sent