Feasibility of an interactive voice response system for monitoring depressive symptoms in a lower-middle income Latin American country

Participants were enrolled between July and October 2014 in three primary care centers in La Paz, Bolivia and its sister city, El Alto. Potential participants were initially identified as part of a 2013 survey of chronic illness care and mobile phone use that was conducted in the same primary care sites [20]. We re-contacted survey respondents in 2014 and invited them to complete a follow-up survey about chronic illness care and mobile phone use. Patients were given an additional option to complete an eligibility screening for two other IVR projects: the current depression study and a study conducted among patients with diabetes or hypertension [27]. Patients eligible for the present study had a PHQ-8 score of 10 or above, indicating at least moderate depression [28]. Participants also had to be between 21 and 80 years of age, have access to a cell or landline telephone, and receive most of their medical care at the clinic where they were recruited. Patients were excluded if they had significant memory problems, significant bipolar symptoms, or a diagnosis of bipolar disorder or schizophrenia.

Eligible patients who agreed to participate completed written consent forms, which a research assistant reviewed out loud with them. These forms described the study purpose and process, and stated that all data collected would be kept confidential and would only be used in the aggregate. All IVR responses were monitored by Bolivian and American research assistants, who followed up with patients who consistently missed calls or had elevated depressive symptoms. Bolivian mental health professionals were available for research assistants to consult with as needed.

Upon enrollment and after informed consent, Spanish-speaking research assistants from the University of Michigan administered baseline surveys to participants to gather data on demographics, mental and physical health and treatments, health behaviors, social support, and health care use. Approximately 1 week after the baseline assessment (mean 7.4 days, range 1–15 days), participants received their first IVR call. The IVR system logged each of the system’s call attempts and completed calls, as well as patients’ touch-tone responses to queries. Follow-up surveys were administered to participants either by telephone or in-person by a research assistant and included closed- and open-ended questions about participants’ satisfaction with the program.

Descriptive statistics were calculated for demographic, health, and depression-related characteristics of program participants. We calculated the proportion of completed weekly calls out of the total number of active call-weeks, and used one-way analysis of variance (ANOVA) to determine whether this proportion varied significantly across groups defined by participants’ age, gender, education, indigenous ethnicity, overall health, and baseline PHQ score.

We used Cronbach’s alpha to assess the reliability of the PHQ-8 administered during IVR calls. We assessed the construct validity of IVR-reported data on depressive symptoms and self-rated health in two ways. First, we sought to determine whether the information patients reported about their symptoms via IVR was consistent with what they told research assistants in face-to-face interviews at baseline. Specifically, we created cross-tabulations to identify the proportion of participants who reported good or better vs. fair/poor overall health in the baseline survey that also fell in these same two categories based on data from the first IVR call (which was closest in time to the survey). We then repeated this cross-tabulation using PHQ score categories <15 (indicating mild-moderate depressive symptoms) vs. ≥15 (severe symptoms) from baseline survey and first call. Next, we used a one-way ANOVA to assess differences in mean self-rated health across completed IVR calls between groups reporting good/better versus fair/poor health at baseline. We repeated this analysis for mean PHQ-8 score across IVR calls and groups with baseline levels of depression that were severe or less than severe, i.e., PHQ-8 <15 vs. ≥15. Finally, we calculated descriptive statistics for participant satisfaction items and identified dominant themes from open-ended responses.

A total of 32 patients with PHQ-8 scores of 10 or higher completed the baseline survey and were enrolled in the study (sample characteristics shown in Table 1). Slightly more than half of participants were 45–64 years of age, and about one-third were over 65. Most participants were women (78 %), non-indigenous (only spoke Spanish at home; 75 %), had completed secondary school (63 %) and were able to read (91 %). As might be expected in a sample of chronically ill patients, no participants reported excellent or very good health at baseline; most reported good (31 %) or fair (56 %) health, and almost half (47 %) reported at least five chronic conditions. Diabetes, hypertension, high cholesterol and chronic back pain were all reported by more than 40 % of respondents.

The mean baseline PHQ-8 score was 13.3 (SD = 3.5). Baseline PHQ scores were significantly higher among men than women (means 16.0 vs. 12.7; p = 0.028), but did not differ significantly by ethnicity, education, self-reported depression diagnosis, self-reported health, number of chronic conditions, or health literacy (not shown in Table). Nearly two-thirds (63 %) of participants screened positive for post-traumatic stress disorders (Table 1). On the Sheehan Disability Scale items (0–10 scale), participants reported disruption levels of 4.1, 3.9, and 4.0 for work, social/leisure, and family/home domains, respectively, indicating moderate depression-related impairment (data not included in Table). Almost all participants (91 %) indicated that they had felt “particularly anxious or nervous” in the last 6 months. Few participants reported receiving clinical treatment for depression, although 16 % reported taking anti-depressant medications, 6 % reported receiving individual therapy, and 6 % reported participating in a support group. However, most reported engaging in positive health behaviors for depression self-care; 69 % reported exercising regularly and 81 % reported eating a healthy diet.

Patients received IVR calls for an average of 12.8 weeks and completed IVR calls an average of 6.9 weeks, for an overall call completion rate of 54 %. Although there were no statistically significant differences in call completion rate among groups defined by demographic and health characteristics, there was a tendency for higher completion rates among: women (56 vs. 44 % of men), those who graduated from secondary school (60 vs. 44 % of those who did not graduate), participants who spoke Spanish vs. an indigenous language at home (58 vs. 39 %), and those with baseline PHQ scores <15 compared to ≥15 (60 vs. 48 %). (data not shown).

Cronbach’s alpha reliability for PHQ-8 scales completed during the IVR assessments was 0.83. When we examined the bi-variate association between patients’ self-reported health status and depressive symptoms at baseline and what they reported in their first IVR call, the Chi square statistics in both cases were non-significant (p = 0.66 for self-rated health and p = 0.60 for depressive symptoms using Fisher’s exact test). However, as shown in Fig. 1, relationships were in the expected direction: among participants reporting good or better health at baseline, 67 % also reported good or better health “today” during the first IVR call. Among individuals with mild/moderate PHQ-8 scores at baseline, 73 % also reported mild/moderate PHQ-8 scores during the first call. Averaged across calls, mean IVR-reported self-rated health was significantly better among patients reporting good or better health at baseline compared to those with fair or poor health (means 2.4 vs. 3.1; F = 5.5; p = 0.03). The mean IVR-reported PHQ-8 score across calls was likewise lower in the group with mild to moderate baseline PHQ-8 scores, compared to those with moderately severe to severe baseline PHQ-8 depression scores, though the difference did not reach statistical significance (means 8.1 vs. 11.3; F = 1.8, p = 0.19).

Nearly all participants (29/30; 97 %) were “mostly” or “very” satisfied with the overall program, as well as with the amount of help they received. Almost three-quarters of participants (77 %) indicated that the program met “most” or “all” of their needs. All participants reported that they would “generally” or “definitely” recommend the program to a friend. Program quality was rated as “excellent” by 37 % of participants, “good” by 50 %, and “fair” by 13 %. Two-thirds (67 %) indicated that the program helped them “a great deal” with managing their depression. In total, 83 % would “definitely” repeat the program; the remaining participants “probably” would (data not shown in Table).

Table 2 shows major themes and example quotations for each open-ended survey item about program satisfaction. Participants noted a number of beneficial aspects of the IVR calls, including: self-care advice; medication adherence reminders; learning that depression was controllable; having someone ask about their health, the ability of the calls to improve mood; and being able to monitor their depression. Participants also identified aspects of their experience needing improvement, including: repetitive questions, technical difficulties (e.g., dropped calls, problems entering touch-tone responses), timing of calls, and preferring human interaction to a machine.

Call completion rates were similar to those in other recent studies of patients with diabetes and hypertension in Bolivia [27]. This is encouraging, because it suggests that despite the hopelessness and passivity that often accompany depression, patients were able to engage in this form of self-care support at a level comparable to a broader group of chronically-ill patients. Also, even with an overall call completion rate of only 54 %, this type of intervention represents a level of patient monitoring and psychoeducation that substantially exceeds what patients currently receive through standard outpatient encounters. On the other hand, the call completion rates that we have observed in Bolivia are lower those we have seen in Honduras and Mexico (roughly 65 %) and among Spanish-speakers in the U.S. (roughly 80 %) [34]. Corresponding rates among English-speaking people with depression in the US are roughly 71 %, and other English-speakers with chronic illnesses complete their calls 85–90 % of the time [35]. It is unknown whether call completion rates as low as the one we observed in this present study represent an adequate “dose” of depression self-management support. Given feedback from some participants about technical difficulties with the calls and dissatisfaction with some aspects of call structure, it is possible that addressing these issues would boost call completion rates.

We found that participants who spoke an indigenous language (Aymara or Quechua) at home—a proxy for ethnicity or degree of ethnic affiliation—had substantially lower call completion rates than participants who spoke Spanish only (58 vs. 39 %). Participants who were more depressed and did not complete secondary school also had notably lower call completion rates; though in this small sample, these differences did not reach statistical significance. Future versions of the program should pay particular attention to the needs of these groups, including translation into indigenous languages, and address their barriers to call completion. Finally, in a recently published pilot trial [36], we found that chronic disease patients in Bolivia who are indigenous, or who have low health literacy or poor medication adherence, are more likely to complete IVR calls when they enroll into the program along with a family caregiver who receives regular feedback about their health and call completion. Other studies from the US and other Latin American countries suggest that providing feedback to family caregivers increases patient engagement as well as program impact upon health and self-management behaviors [29, 35, 37‐39]. Broadening the current depression self-care support program to include family members may therefore enhance patient engagement and outcomes.

The internal consistency reliability (Cronbach’s alpha) of the PHQ-8 collected in IVR calls was good at 0.83, and almost identical to a previous coefficient of 0.84 in an unpublished analysis by the authors of data from an in-person survey delivered in a sample of 600 Bolivian primary care patients. Researchers elsewhere in Andean Latin America have confirmed the scale’s cultural appropriateness [40]. We assessed the validity of IVR-collected self-rated health and depressive symptom data by testing whether it related in expected ways to the survey data, which we assumed was more accurate as it was collected using a well-established standard method. These associations were in the expected directions, although they did not always reach statistical significance. We note that the association between baseline survey- and IVR-collected data of baseline PHQ-8 and self-rated health scores would naturally be attenuated by the fluctuation of depression symptoms over time. The self-care messages included in the calls may also have alleviated depressive symptoms in some cases. Nonetheless, the patterns we observed in our data were consistent with our hypothesis that patients in the Bolivian primary care system with elevated depressive symptoms can accurately report their health and mood status via IVR calls.

Over three-quarters of patients in our sample were female. Across countries and cultures, women are more likely to be depressed than men [41, 42]. Notably, the men in our sample had significantly higher baseline PHQ scores, possibly reflecting that women are willing to participate in mental health programs at a lower level of symptomatology. Although the mean overall baseline PHQ score for the sample (13.3) fell near the center of the “moderate” range of depressive symptom severity, only one-third of participants reported having a depression diagnosis. As expected given the scarcity of mental health workers and psychotropic medications in Bolivia, as in other LMICs [10, 43] as well as a weak infrastructure for identifying and treating primary care patients who have mental health disorders [44], depressive symptoms appeared to be virtually untreated in our sample. Very few participants reported taking part in either counseling or support groups for depression, or taking antidepressant medications. In contrast, a large majority reported either exercising regularly or following a healthy diet to self-manage depression. This fact demonstrates a willingness to engage in self-care for depression in the form of positive health behaviors, as observed also in participants’ responses to open-ended questions. Finally, our participants tended to report only low to moderate functional impairment due to depression. Individuals with higher levels of functional impairment may be unlikely to participate in an IVR-based self-care program.

Comorbid mental health disorders appeared to be highly prevalent in our sample, with nearly two-thirds of participants screening positive for post-traumatic stress disorder and over 90 % indicating possible anxiety disorder (i.e., they had been particularly anxious or nervous in the last 6 months). Comorbidity between psychiatric disorders is common in LMICs, and in these settings it is often impractical to administer separate evidence-based interventions for co-occurring psychiatric illnesses [45]. The common elements treatment approach (CETA) developed by Murray and colleagues for delivery by non-specialists incorporates flexible treatment elements targeting a wide range of psychological symptoms. Because CETA involves weekly symptom monitoring to inform treatment element selection and dose, this approach may be well-suited for incorporation into IVR or other mHealth interventions to treat a variety of common, and/or comorbid, mental health disorders.

Some participants (Table 2) revealed in open-ended comments that they valued the attention, guidance, and feedback that they received as a result of participating in the program. Some comments implied at least a temporary therapeutic effect of the IVR contact; for example: “It was motivating to have someone call and be concerned about my health, and no matter how sad I was feeling, the advice they gave me always cheered me up.” The messages regarding self-care and taking control of depression seemed to resonate with participants, even in a culture that is often seen as fatalistic and passive: “One learns to trust oneself…and take control of this illness.”

Nonetheless, responses also suggested areas for program improvements. A number of participants felt that the program would be more compelling if it incorporated a “human element.” “Task-shifting” in the form of training lay and non-specialist health workers to deliver brief, structured psychological treatments under specialist supervision is an evidence-based approach recommended by the World Health Organization for delivering mental health care in LMICs [46, 47]. Psychotherapeutic interventions that combine electronically-delivered elements with contact with a non-specialist health worker are growing in use [48] and may be a low-cost way to augment mHealth interventions with personal contact, while at the same time leveraging the structure and consistency of the automated electronic components to increase the capacity of lay health workers to provide high-quality psychotherapy. In Bolivia, a manual for delivering cognitive-behavioral therapy that is based on WHO guidelines for depression care and also tailored to the specific socio-cultural context of Bolivia is under development; this is an important resource that will facilitate training local health workers (Dra. Rosa Tarrazona, personal communication, January 26, 2016).

Technical difficulties during the IVR calls, such as problems entering data or completing their calls, were frustrating to participants. Aside from addressing this problem with improved technology, if available, live technical support would have helped participants resolve problems with their phone or the IVR system, and may reduce program attrition. Last, one participant alluded to need to address the role of intimate partner violence: “[Maybe there should be] advice for couples, because there are many women who are beaten by their husbands and they get depressed and they need help getting out of this problem.” (Quotation not shown in Table 2.) This comment, along with data indicating that nearly half of all Bolivian women reported experiencing violence from their intimate partner in the last year [49], point to the importance of addressing social and safety issues as part of the service, including a mechanism for referrals to appropriate resources.