Self-help mobile messaging intervention for depression among older adults in resource-limited settings: a randomized controlled trial

Study design and participants

PRODIGITAL-D was a pragmatic, single-blind, two-arm individually RCT with 1:1 allocation, conducted in 24 primary care clinics, also called UBSs, in underprivileged areas of the city of Guarulhos, Brazil. This city is part of the metropolitan region of São Paulo with a population of around 1.3 million.

We recruited individuals aged 60+ years, registered with any of the participating 24 UBSs, able to receive WhatsApp messages, and who screened positive for depressive symptomatology (PHQ-9 ≥ 10)²³ with at least one core symptom of depression (low mood or anhedonia, that is, PHQ-2 ≥ 1)²⁵. The PHQ-9 threshold had a sensitivity of 91% and specificity of 88% against diagnostic interviews for adults aged 60+ years²⁴. Exclusion criteria included substantial visual or hearing impairments that would hinder comprehension of messages on a mobile phone, acute suicidal risk identified during the screening assessment, another individual residing in the same household already participating in PRODIGITAL-D or previous participation in the PROACTIVE trial. Informed verbal consent was obtained before the screening assessment and when participants were invited to the trial, both conducted by phone. Consent was audio-recorded after authorization from the older adult. Details of the protocol are published elsewhere¹⁷.

This study was approved by the ethics committee of the Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo (Comissão para Análise de Projetos de Pesquisa, ref: 4.097.596, first approved 10 March 2021) and authorized by the Guarulhos Health Secretary. It was registered with the Brazilian Registry of Clinical Trials (ReBEC), RBR-4c94dtn, on 22 October 2021. Trial registration materials were submitted to ReBEC on 3 August 2021.

Procedures

A list with all the names and phone numbers of older adults registered with the 24 participating UBSs was provided by the Guarulhos Health Secretariat. After excluding duplicates and individuals without phone numbers, individuals were assigned a random ID for our study, as the original list was in alphabetical order. Research assistants then sent a WhatsApp message to older adults according to the new list sorted by ascending random ID numbers to search for individuals with a valid mobile number and who used WhatsApp. Individuals with a valid WhatsApp number (that is, who received the prescreening WhatsApp message) were approached by telephone to be screened further for depressive symptomatology (using the PHQ-2)²⁵. Up to two call attempts were made on different days. Individuals with a PHQ-2 score of one or greater were then asked the other seven items of the PHQ-9 questionnaire²³. Those with PHQ-9 scores of ten or greater completed a baseline assessment to ascertain the exclusion criteria; eligible individuals were invited to participate in the trial. The screening and baseline assessments together lasted approximately 35 min. The invitation for the trial, which includes reading the full participant information sheet, occurred in the same call, whenever the individual agreed to continue it for around 15 min longer. When this was not possible, an additional call no more than 28 days after the PHQ-9 screening was made to complete the recruitment. Follow-up data were collected by phone at 3 (weeks 12–16) and 5 months (weeks 20–24) after sending the first message (intervention) or the single message (control). A 4-week window was used to maximize response to follow-up considering the logistical realities of conducting phone interviews.

Quality control of a random sample of recorded assessments was conducted by an independent research assistant (not involved in any data collection) to ensure all procedures had been followed and the quality of data collected.

Randomization and masking

Participants were stratified according to age group (60–69 years, 70–79 years, 80+ years), sex (male, female) and severity of depressive symptoms (baseline PHQ-9 categories: 10–14, 15–19, 20+). The allocation sequence was generated using randomly permuted blocks with random block sizes of six, eight or ten by research team members not involved in data collection (C.A.N. and T.J.P.). The ‘randomization module’ of the Research Electronic Data Capture (REDCap)^26,27 was used to conceal the allocation sequence and randomize individuals.

Research assistants involved in recruitment and follow-up data collection were blinded to trial allocation. Two different groups of research assistants were responsible for either the first or the second follow-up assessment. Researchers responsible for scheduling the automated delivery of messages were not blinded to trial allocation but had no role in collecting the inclusion or outcome data. Due to the nature of the interventions, it was not possible to mask trial participants.

Interventions

Participants in both arms continued receiving their UBS’s usual primary care. The research team did not interfere with that care (consultation or treatments, including medication) and the UBS health professionals had no role in Viva Vida.

During the first 2 weeks (for the intervention arm) and after the single message (for the control arm), a technical support person identified and contacted individuals who were having issues with receiving or opening the message. This was to help participants with any potential technical issues that were preventing them from accessing the messages.

Viva Vida program

Participants allocated to the intervention received a psychosocial intervention delivered via WhatsApp. A total of 48 audio or visual messages were automatically sent 4 days a week for 6 weeks (one in the morning, one in the afternoon). The audio messages were based on psychoeducation²⁸ or behavioral activation²⁹ principles and lasted on average 3 min using storytelling techniques. In the audio messages, there were two characters (Ms Ana and Mr Leo) reading and commenting on letters sent by fictional older adults participating in the program. Through these letters they shared stories about their lives and symptoms of depression and how the Viva Vida intervention helped them to feel better. Eight visual messages including summary diagrams were used to reinforce the content of the audio messages. Additionally, at the end of each week, participants received an extra message with a question about their opinion of the program that could be responded to using the WhatsApp quick reply tool. Participants also received one message at the beginning of Viva Vida with a support contact number in case they experienced technical issues receiving the messages. As the messages were delivered automatically, this strategy allowed participants to contact us in case they were not receiving the intervention as intended. During the program, participants were also advised to visit the UBS if they needed professional support for any health issues. Extended Data Fig. 1 shows the structure of the Viva Vida program and the content of the audio and visual messages that participants received each week.

The Viva Vida messages were based on the PROACTIVE intervention delivered to older adults^10,18,30. The use of WhatsApp as a tool to deliver an intervention was initially discussed in a group with six older adults and one community health worker in one UBS. After developing the messages, selected older adults received a sample and were interviewed to evaluate qualitatively the acceptability of the messages. The automated system and the recruitment by phone were subsequently evaluated in a feasibility study with a random sample of older adults with depressive symptomatology recruited from two UBSs in Guarulhos that did not participate in the study.

Single message control arm

Control arm participants were sent a single and brief (6-min) audio message providing information about depression and simple ways to deal with depressive symptoms. It also offered additional tips encouraging participants to live a healthier life, such as improving sleep and maintaining a balanced diet, advising them to seek health care if they felt they needed additional support.

Outcomes

Improvement from depressive symptomatology at 3 months was the primary outcome. We considered improvement when the PHQ-9 scores were below 10 (ref. ²³). Secondary outcomes included improvement from depressive symptomatology at 5 months and reduction in PHQ-9 scores by at least 50% between baseline and follow-up visits at 3 and 5 months. We also evaluated as exploratory outcomes the effects on the continuous scores of depressive symptomatology (PHQ-9), anxiety symptomatology with the GAD-7 (ref. ³¹), loneliness with the 3-item UCLA loneliness scale³², health-related quality of life EQ-5D-5L³³ and capability well-being with the ICECAP-O³⁴.

Safety

As mentioned before, participants continued receiving usual primary care and were advised to contact the primary care team whenever they felt they needed additional support from health professionals. The risk associated with receiving the interventions was considered minimal. Severe adverse events, including acute suicidal risk, hospitalizations and death were collected during the follow-up assessments using a standardized form. Participants who reported suicidal ideation were assessed for suicidal risk with the standardized protocol used in a previous study⁷. UBS managers were contacted to inform about participants with suicidal ideation. We also investigated any hospitalization or death with participants or family members to understand if it was related to study participation.

Data collection and management

All the measures mentioned above were applied at baseline, and at the 3-month and 5-month follow-up assessments. Sociodemographic and clinical characteristics were also collected at baseline. Economic data were collected at each follow-up assessment. Severe adverse events were assessed for all participants at 3 and 5 months.

Data were collected and managed using REDCap^26,27 hosted at the Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo. REDCap is a secure, web-based software platform designed to support data capture for research studies. A system was developed to collect and store data from the WhatsApp messages. A dedicated server component scheduled the messages for groups of participants (grouped according to trial arm and the start date of the messages based on the date of randomization) and pushed these to the participants using the WhatsApp Business interface. The latter provides time stamps indicating when participants receive and open their messages, as well as the answers chosen using the quick reply tool and the content of spontaneous messages sent by participants.

Sample size

With an assumed attrition of 25%, 440–500 randomized individuals would yield 80–85% power to detect a 15 percentage point difference in depression improvement (PHQ-9 < 10) rates between the control and intervention arms at 3 months (25% versus 40%) using a two-sided 5% alpha. Such a difference in improvement rate is considered clinically meaningful^8,30 but there were no local studies that had investigated a clinically meaningful threshold.

Statistical analysis

Descriptive statistics were obtained using complete cases. All comparative analyses were performed using intention-to-treat principles with imputed data in regression models, aligned to the SAP contingent on the amount of missing data³⁵. Logistic regression was used for binary outcomes; linear regression was used for continuous outcomes. All regression analyses were adjusted for stratification, and all analyses with an outcome other than PHQ-9 were also adjusted for the baseline score of the corresponding outcome.

Prespecified subgroup analyses at both follow-up assessments were investigated using likelihood ratio tests for interactions between randomized arm and the following: sex; age; educational level; comorbid physical illness (hypertension, diabetes or both); and baseline PHQ-9 severity categories. We also ran a model for the primary outcome adjusting for elapsed time between consenting to participate in the trial and the first or primary follow-up.

Initially, patterns of missingness were investigated by comparing missing with complete data for the baseline demographic characteristics, according to trial arm, at both follow-up visits. To reduce bias and loss of information, we used MICE with 50 imputations, as implemented in the multiple imputation command in Stata under the assumption that data were MAR³⁶. Variables included in the MICE models consist of the ‘improved from depression’ outcome, covariates described in the SAP (age group, severity of PHQ-9, sex, treatment arm) and variables found to be predictors of missingness^37,38. Predictors of missingness included the following: hypertension at baseline, income and levels of enjoyable activities participants engaged in. Also included were baseline scores for corresponding outcome models, including depressive symptomatology (PHQ-9), anxiety symptomatology (GAD-7), health-related quality of life (EQ-5D-5L), capability well-being (ICECAP-O) and loneliness (3-item UCLA).

To assess the sensitivity of our findings against modest departures from the MAR assumption, a weighted sensitivity analysis using the selection model approach was applied^39,40,41. Briefly, once data had been imputed under MAR, parameter estimates from each imputed dataset were reweighted to allow for the data to be missing not at random. To test the stability of our model, we considered different degrees of departure from the MAR assumption by considering plausible values of δ ranging from 0.10 to 0.40. This range corresponds to ORs for the data being observed when a participant improved from depression compared to when they did not, ranging from 1.11 to 1.50 (that is, the exponential of 0.10 and 0.40 respectively).

CACE analysis⁴² using an instrumental variable estimator and imputed data was applied to estimate the effect of the number of messages electronically recorded as being opened on depression outcomes. The protocol¹⁷ and the SAP³⁵ prespecified the ‘minimum therapeutic dose’ as listening to ‘most of the messages’ and ‘more than half of the messages’ received, respectively because it was initially based on self-reported categories (‘none’, ‘a few’, ‘at least half’, ‘most of the messages’ and ‘all of the messages’) of a question from the first follow-up assessment. Because in the event we were able to access the electronic data, we operationalized the prespecified minimum therapeutic dose as opening 36 or more messages (75% of the total of 48 messages) versus opening 35 or fewer messages. The CACE analyses were conducted using the PHQ-9 score at 3 and 5 months, adjusting for stratification. We also conducted sensitivity analyses using thresholds of opening at least half of the total messages (24+) versus 23 messages or fewer, and opening all 48 messages versus not doing so.

Regression diagnostics were run for all regression models. The Box–Tidwell test was run after the logistic regression models to test whether the logit transform was a linear function of the predictors for the different models. Normality assumptions for linear regression models were evaluated through residual plots. Statistical tests were two-sided and all analyses were conducted using Stata v.17 (StataCorp LLC).

Ethics and inclusion statement

Most of the authors (10 of 15) are from Brazil, where the RCT took place, and were funded to work on the development and delivery of the intervention, data collection or study management. The authors based in high-income countries are researchers who contributed to the conception of the study and supported the application to obtain funding shared between Brazil and the UK. This study is relevant for Brazil and other LMICs. One member of the trial steering committee is from the Guarulhos Health Secretary’s office, while two other members are researchers based in other cities of Brazil. Meetings with primary care and UBS managers in Guarulhos were held before, during and after the trial. This study was approved by the ethics committee of the Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo (Comissão para Análise de Projetos de Pesquisa, ref. 4.097.596) and authorized by the Guarulhos Health Secretary.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.