Network analysis of mental health problems among adults in Addis Ababa, Ethiopia: a community-based study during the COVID-19 pandemic

STRENGTHS AND LIMITATIONS OF THIS STUDY

  • The sample size is large (n=1127), which adds to our confidence in the precision of the estimates.

  • We have included participants from subcities with high, medium and low reported cases of COVID-19 in Addis Ababa to control the effects of case loads on mental health.

  • Symptom-level analysis of mental health problems is not common in low-income countries, and this study could be used as a reference for similar studies from low-income countries.

  • Using network analysis to examine the comorbid depression, anxiety and perceived stress symptoms among adults in Addis Ababa.

  • The network analysis was based on a cross-sectional study design using undirected network analysis, which does not establish causality links related to a symptom and does not show changes over time.

Introduction

The COVID-19 caused by SARS-CoV-2 is a devastating global pandemic that resulted in millions of infections and nearly 7 million deaths until late December 2023. Though the COVID-19 pandemic has affected all sectors, the crisis was worse in many countries’ already overburdened health systems. Evidence indicates that the COVID-19 pandemic has caused significant disruption in health service delivery, particularly in resource-limited countries (impact). Among its different impacts, the COVID-19 pandemic has worsened short-term and long-term stress and compromised the mental health of millions of people worldwide, resulting in a global mental health catastrophe.1 Mental disorders are among the top 10 leading causes of disease burden globally, with no evidence of a reduction in the past several decades, and the emergence of the COVID-19 pandemic created an environment where many determinants of poor mental health outcomes have been exacerbated.2 3 The reported prevalence estimates of psychological disorders during the pandemic are higher than in the prepandemic periods.4 5 Different evidence has indicated that increased psychosocial distress in the general population is due to the nature of the disease, loss of loved ones and the wide range of responses that were implemented to prevent and control the spread of the infection, including strict infection control, quarantine, physical distancing and lockdowns.4 6 7 Additionally, mental health problems were linked to direct or indirect effects of the pandemic.8 Direct effects included fear of contagion or worries about becoming infected and perception of the risk of dying.4 9 The indirect effects were due to economic and social circumstances such as lifestyle changes and worsening living conditions.4 10

According to the WHO mental health report, a greater proportion (82%) of the people suffering from mental health disorders (MHD) reside in low and middle-income countries (LMICs).1 This has been worsened by the mental health impacts of the COVID-19 pandemic, which have been disproportionate in LMICs. The pandemic’s strong psychological effect in these settings is due to the pre-existing mental health risk factors, severe financial constraints, limited access to healthcare, inadequate health insurance coverage and uncertainty brought on by a lack of pertinent information. The increased mental distress in Africa was not appropriately addressed due to the weak health system.11 Similarly, an elevated level of depressive and anxiety symptoms during the first year of the COVID-19 pandemic has been reported by different studies conducted in Addis Ababa and elsewhere in the country. According to a recent meta-analysis conducted in Ethiopia, the pooled prevalence of the psychological impact of COVID-19 in Ethiopia was 42.50%, and subgroup analysis of the same study indicated a higher prevalence (66.4%) in Addis Ababa.12 Early identification of individuals in the initial stages of a psychological disorder increases the effectiveness of intervention strategies.9 However, diagnosing a mental health condition is difficult as some symptoms that belong to one disorder are also symptoms of other disorders.13 Mental health symptoms commonly co-occur in the same person, indicating the presence of one mental disorder acts as a trigger for the other, resulting in the comorbidity of MHD.13–15 Compared with having one mental disorder, the co-occurrence of two or more MHDs is associated with more severe psychopathology and poorer clinical outcomes.13 16 Comorbidity of depression and anxiety is associated with more severe health outcomes such as greater illness severity, higher risk of chronicity and more severe functioning impairment.17–19 Other evidence also indicated higher COVID-19-related perceived stress predicted an increased risk of anxiety and major depressive disorder and comorbid anxiety and depression.20

Researchers and mental health professionals use several tools to assess the mental health of a population for different purposes, which include screening for the presence or absence of a mental health condition, making a formal diagnosis or monitoring progress. Researchers commonly use a battery of brief MHD screening questionnaires to measure common mental disorders 21 . Accordingly, most of the studies conducted during the pandemic used standard and validated scales to screen for MHD such as depression, anxiety and comorbidities.22 23 Such screenings or analyses rely on individual or sum scores and do not consider the direct interactions between the symptoms, providing critical information on mental disorders management.24 Due to this limitation, scale-based measurements of mental health offer insufficient evidence for identifying psychopathological mechanisms of mental disorders.25 In contrast, network analysis has gained substantial attention as an alternative model to establish causal relations between symptoms and identify marker symptoms by quantifying the relationships between individual mental health symptoms.22 24–26 According to network analysis, symptoms do not arise from potential causes but rather are inter-related, causally interconnected and reinforce each other, eventually leading to the establishment of whole mental disorders.13 A study that compared mental health outcomes using individual scores versus symptom networks revealed different outcomes, with network analysis dimming the positive results gained from individual scores analysis.24

Poor mental health is associated with multiple functional impairments. Conventional methods of studying the psychopathology of mental health have implicitly assumed that the symptoms are signs of an underlying illness. Opposed to this, the network approach to psychopathology assumes mental disorder is made up of co-occurring symptoms and their propensity to reinforce one another causally.27 Understanding the mechanisms to reduce the risk of severe health outcomes and provide timely and effective treatments is essential. However, network analysis studies of associations between symptoms of common mental health syndromes during the pandemic are lacking in low-income countries. It is of significant relevance to understand how symptoms relate to each other, which could help identify mental health problems early and initiate appropriate interventions.28 To our knowledge, network analysis has not been done in Ethiopia and other sub-Saharan African countries to examine symptom-level interaction of MHD. Therefore, this study aimed to identify the central and bridge symptoms in the network structure of depression, anxiety and perceived stress among the general adult population of Addis Ababa, Ethiopia.

Methods

Study setting and design

This survey was conducted among the adult population of Addis Ababa, the capital city of Ethiopia, during the first year of the COVID-19 pandemic. Addis Ababa is an urban setting with an estimated 5.2 million people.29 The capital had the highest burden of confirmed cases and death from COVID-19 in the country.30 The country’s largest COVID-19 treatment centres and hospitals were also in the city. A community-based cross-sectional study was carried out among the general adult population of Addis Ababa. The inclusion criteria were 18 years and above and a permanent resident of the study subcities. A total of 1127 participants were included in this study, of which 747 (66.3%) were females and 380 (33.7%) were males.

Data collection and measurements

The household survey was conducted in three selected subcities of Addis Ababa in September 2020 and 2021. This allowed identifying a minimum of 1500 households. Adults identified by simple random sampling were included in the study and called on their cell phones to conduct the survey. Finally, 1127 complete data were included in the analysis. Data were captured into a mobile tablet-based data collection system and promptly uploaded to a secure server after collection. Open Data Kit was used for the data collection. Trained data collectors with previous experience in phone surveys have collected the data. All 10 data collectors have bachelor’s degree in health and related fields.

Depression

The Patient Health Questionnaire 9 (PHQ-9) was used to measure depressive symptoms. The tool has nine items to inquire about depressive symptoms over the past 2 weeks, as indicated in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) manual.31 All nine items were rated from ‘0’ (not at all) to ‘3’ (nearly every day). Total scores ranged from 0 to 27, with higher scores indicating more severe depressive symptoms. Having elevated depressive symptoms was defined by a PHQ-9 score of 10 or greater.32 The PHQ-9 has been validated as an efficient tool for screening depression among the adult population in Ethiopia.33 The Cronbach’s alpha for the PHQ-9 was 0.73.

Anxiety

Generalized Anxiety Disorder Scale (GAD-7), a validated tool, was used to measure the severity of anxiety symptoms.34 35 Each of the GAD-7 items assessed common anxiety symptoms in the past 2 weeks. The items were rated from 0 (not at all) to 3 (nearly every day); total scores ranged from 0 to 21. A total score of >10 points is considered the cut-off for classifying as having anxiety.36 The GAD-7 has a Cronbach’s alpha of 0.71 in this study.

Stress

Perceived Stress Scale 10 (PSS-10) was used to measure the perception of stress related to the pandemic. PSS-10 is a 5-point Likert scale from 0 (never) to 4 (very often), yielding a score ranging from 0 to 50.37 To calculate a total score of PSS, responses to the four positively stated items (items 4, 5, 7 and 8) were reversed (ie, 0 to 4; 1 to 3; 2 to 2; 3 to 1; 4 to 0). A score of >20 points was considered the cut-off for experiencing perceived stress on COVID-19. PSS (α=0.71) is validated for use in Ethiopia.38

Statistical analysis

A descriptive analysis was done to describe the participant’s profile, and mean values were calculated for continuous variables, whereas frequency and percentage were calculated for categorical variables. The R program V.4.2.3 was used for data analyses.

Network estimation

Skewness and kurtosis were done to assess the distribution of the items. All items of PHQ-9 (skewness=2.1, kurtosis=7.5), GAD-7 (skewness=1.7, kurtosis=5.6) and PSS-10 (skewness=0.3, kurtosis=3.1) had a normal distribution. Hence, network estimation was done using the Gaussian graphical model, and graphical least absolute shrinkage and selection operator technique26 and extended Bayesian information criterion were used to establish the depression-anxiety-perceived stress symptom network structure. These methods were used to shrink edges in the network and tune parameters to make the symptom network sparser and easier to interpret.22 26 R packages graph (version 1.9.4) and bootnet (version 1.5) were applied to estimate and illustrate the network model visually. A node represents a variable or symptom in the network analysis, while an edge is a pairwise association between variables or symptoms. Partial correlation analyses were done to build the association of each pairwise continuous variable (nodes) and form a network. The thickness of an edge represents the strength of an association between two nodes, and while the colour indicates the direction of the association, blue and red edges reveal positive and negative correlations, respectively. In correlation-based networks, an edge with a positive value indicates that an increase in the activation of one node is associated with an increase in the activation of the node connected to it. In contrast, a negative edge indicates an increase in the first node is associated with a decrease in the second node.26 39 The predictability of each node in the network was assessed using the R package mgm (version 1.2-13). Predictability is the extent to which the variance of a node is explained by adjacent nodes in the network.

Estimation of network accuracy and stability

The robustness of the network structure was evaluated by estimating the accuracy of the edge weights by computing CIs with a non-parametric bootstrapping method. Additionally, bootnet based on 1000 bootstraps was performed for each node to assess the stability of the centrality index.26

After reviewing the network structure, the expected influence (EI) index was calculated to identify the most central symptoms. EI is a more appropriate measure of centrality to predict node influence on a network containing both positive and negative edges.40 Nodes with higher EI in the network mode were considered more important. Furthermore, to explore bridge symptoms in the network that played essential roles in connecting two or more psychiatric disorders, the bridge EI (1-step) was calculated using the R package networktools (version 1.5.0).26 39 Nodes having higher bridge EI values reflected a greater risk of contagion from one community to other communities compared with bridges with lower bridge EI values.39 The centrality indexes of EI and bridge EI were reported as standardised values (Z scores) and considered stable when the correlation stability coefficient (CS-C) was larger than 0.25 and preferably larger than 0.50.26

Network comparison tests

A network comparison test (NCT) with 1000 bootstraps was done using the R package Network Comparison Test (version 2.2.1) by considering the moderating effects of gender, age and survey time.

Patient and public involvement

No patient was involved.

Results

A total of 1127 participated in this study. About one-fourth attained tertiary education/college or above. Of the total, 27.59% were wage employees, and more than half (57.7%) were unmarried/never lived together . Most (77.8%) participants reported themselves as head of the household (table 1). The prevalence of having any mental health problems was 10.7% (95% CI 8.97% to 12.6%).

Table 1

Background information of study participants in Addis Ababa, Ethiopia, 2021

Network structure

All items used in measuring depression, anxiety and perceived stress were included, and no item was excluded for redundancy. The network structure of depressive, anxiety and perceived stress symptoms has both positive and negative edges, as shown in figure 1. The network model indicated that anxiety has the strongest edge in the network, the connection between GAD-1 (nervousness) and GAD-2 (uncontrollable worry) (weight=0.36). In the perceived stress community, stronger connection was observed between PSS 9 (Being angered) and PSS 10 (Could not overcome) (weight=0.28), and between PSS 4 (Felt confident) and PSS 5 (Going your way) (weight=0.28). The association of PHQ-8 (motor) with PHQ-9 (suicide) (weight=0.27), PHQ-3 (sleep problem) with PHQ-5 (appetite) (weight=0.23) and PHQ-1 (anhedonia) with PHQ-2 (sad mood) (weight=0.21) has stronger edges within the depression. The association between PHQ-2 (sad mood) and GAD-1 (nervousness), PSS-2 (unable to control) and PHQ-6 (guilt), and GAD-3 (excessive worry) and PSS-3 (nervous and stressed) showed a stronger association across the depression and anxiety, depression and perceived stress, and anxiety and stress, respectively. The mean predictability was 0.25, indicating that, on average, neighbouring nodes could account for 25% of each node’s variance (table 2). Three symptoms, PSS-3 (nervous and stressed), PHQ-2 (sad mood) and PHQ-6 (guilt), had the highest predictability in the model. On the opposite end, GAD-5 (restlessness), PSS-2 (unable to control) and GAD-4 (trouble relaxing) had the lowest predictability (table 2).

Figure 1
Figure 1

The network structure of depression, anxiety and perceived stress symptoms among adults in Addis Ababa, Ethiopia, 2021. GAD, Generalized Anxiety Disorder Scale; PHQ, Patient Health Questionnaire; PSS, Perceived Stress Scale.

Table 2

Descriptive statistics of the depression, anxiety and perceived stress measurement items in Addis Ababa, Ethiopia, 2021

Central symptoms and bridge symptoms

The centrality index EI indicated that node PHQ-2 (sad mood) (EI=1.52) had the highest EI value, followed by the nodes PHQ-6 (guilt) (EI=1.37) and PSS-3 (nervous and stressed) (EI=1.36), as shown in the network structure (figure 2). This indicated that these three symptoms were central symptoms influential for understanding the structure of the depression, anxiety and perceived stress network model among adults in Addis Ababa. The statistical difference between the EIs and bootstrapped edge CIs can be looked at in online supplemental figures 1 and 2.

Supplemental material

Figure 2
Figure 2

Node expected influence of the depression, anxiety and perceived stress network structure, Addis Ababa, Ethiopia, 2021. GAD, Generalized Anxiety Disorder Scale; PHQ, Patient Health Questionnaire; PSS, Perceived Stress Scale.

The bridge EI index showed PSS-3 (nervous and stressed) bridge EI=1.33, GAD-6 (irritability) bridge EI=1.12 and PHQ-2 (sad mood) bridge EI=1.10 were the bridge symptoms linking depression, anxiety and perceived stress communities. This indicates that these three symptoms can increase the risk of contagion to anxiety, depression and perceived stress.

Network stability and accuracy

The depression-anxiety-perceived stress network showed good stability. Node EI’s CS-C was 0.67, indicating that 67% of the sample could be dropped and the network structure would not significantly change (figure 3). For bridge EI, the CS-C was also 0.67, which was also good.

Figure 3
Figure 3

Stability of the bridge node expected influence by case-dropping subset bootstrap, Addis Ababa, Ethiopia, 2021.

Network comparison tests

An NCT was conducted to compare differences across genders in the depression-anxiety-perceived stress network model. However, the comparison of network models between female and male participants did not show significant differences in network global strengths (9.18 in 10.61 male and female participants, respectively; S=1.42, p=0.196) and edge weights (M=0.205, p=0.571). Additionally, the network models were compared between the younger and older age groups. No difference was observed between younger22–41 and older age group (41 and above) participants (p value 0.67).

Discussion

To the best of our knowledge, this was the first study using network analysis that documented the network structure of depression, anxiety and perceived stress in sub-Saharan African countries, including Ethiopia. This study identified the most central and bridge symptoms linking depression, anxiety and perceived stress disorders in the network structure of the general adult population of Addis Ababa during the first year of the COVID-19 pandemic. The central symptoms in the network were PHQ-2 (sad mood), which had the highest EI value, followed by the nodes PHQ-6 (guilt) and PSS-3 (nervous and stressed). Meanwhile, nodes PSS-3 (nervous and stressed), GAD-6 (irritability) and PHQ-2 (sad mood) were the bridge symptoms and had the highest bridge EI. This study found that the strongest edges were within a particular community rather than between different mental disorders, and the edges in the anxiety community were the strongest. The results of the network analysis were robust to stability and accuracy tests.

Several circumstances contribute to the development of depression, including stressful life events or difficult experiences. Events such as the death of a loved one and isolation trigger loneliness and mood changes, which are associated with feelings of emptiness and sadness.41 A sad mood is the hallmark symptom required for meeting the diagnosis of major depression.42 The emergence of COVID-19 necessitated implementing or enforcing different public health safety measures, including isolation or quarantine, and such measures are associated with social isolation and loneliness, which are stronger predictors of mental health problems.41 Psychological distress was intensified among individuals who experienced sad moods because of the loss of loved ones to COVID-19.28 43

Strict isolation regulations during the pandemic restricted families from attending events, including burial processes of loved ones deceased due to COVID-19-related illnesses. In our study, the sad mood had the highest centrality and bridge EI. This is consistent with other studies that identified sad mood as a central and bridge symptom linking depression with anxiety and other mental disorders.22 42 44 Similar to other studies, the edge between PHQ-8 (motor) and PHQ-9 (suicide) was the strongest in the depression community.14 Psychomotor agitation is the most frequently reported symptom among persons who reported previous suicide attempts.14 This result showed the importance of giving more attention to individuals with specific symptoms of depression to minimise catastrophic consequences.

Anxiety had the strongest edge in the network, which is between GAD-1 and GAD-2. Irritability is a bridge symptom linking anxiety with other disorders. This means that irritability has the strongest ability to increase the risk of contagion to depression and perceived stress disorders in the current network structure. This finding is consistent with other network analysis studies that used GAD-7 to measure anxiety among adults.13 22 Irritability is characterised by feelings of anger or frustration, and DSM-IV states that irritability serves as a diagnostic criterion for generalised anxiety disorder and depression.31 Higher irritability levels resulted in a more severe psychological impact during COVID-19. Those who suffer more severe consequences of the pandemic experience more irritability as a direct result of those consequences on family members and/or household economies. The population in low-income countries is disproportionately affected by poverty, further compromised by the pandemic.45 46 Similarly, the pandemic worsened Ethiopia’s unemployment rates, household incomes and food insecurity, increasing irritability.47 48

Changes in social environments such as social distancing and lockdown, worrying about infecting families, unexpected bereavements and uncertainty about the future have increased nervousness and stress. Stress is one of the primary risk factors for MHD, which was exacerbated by the COVID-19 pandemic. Practising precautionary measures were reported to be strongly correlated with perceived stress.49 50 This study showed that those who reported perceived stress had stronger edge weights, which are PSS-9 (being angered) and PSS-10 (could not overcome). Furthermore, the network analysis revealed nervousness and stress as bridge symptoms linking perceived stress with anxiety and depression disorders.50

Depression, anxiety and stress are common MHD. Symptoms of one of these mental conditions increase the chances of getting other mental disorders, leading to diagnostic comorbidity. The findings of this network analysis indicated that the bridge symptoms of different mental disorders contribute to increased psychiatric comorbidity. Our results identified sadness, irritability, nervousness and stress as bridge symptoms. People with comorbidity experience more severe psychosomatic symptoms, physical and mental inability and lower quality of life than those with one disorder.13 16 Identifying mental health problems solely relying on the sum or mean scoring that gives equal weight to all symptoms may be misleading when determining specific symptoms underlying comorbidities.

Strengthening access to psychological counselling is very important to minimise the risk of developing mental health problems either separately or as a comorbidity. However, access to mental healthcare is complicated by poverty, weak infrastructure, lack of trained manpower and limited access to mental healthcare facilities. The central and bridge symptoms we identified in the current network model could be targeted in the treatment and preventive strategies among the general adult population.

Strengths and limitations of the study

The strengths of this study included using network analysis to examine the comorbid depression, anxiety and perceived stress symptoms among adults in Addis Ababa. To the best of our knowledge, this was the first study to document depression-anxiety-perceived stress symptom networks. The study also has limitations. First, our network analysis was based on a cross-sectional study design using undirected network analysis, which does not establish causality links related to a symptom and does not show changes over time. Second, self-report measures of symptoms were used, which could introduce recall bias and social desirability bias. Third, the study focused on adults, and the findings could not be generalised to other groups. However, we suggest replicating it in adolescents and young adults.

Conclusion and implication

This study assessed the network structure of depression, anxiety and perceived stress symptoms among adults in Addis Ababa during the COVID-19 pandemic. Sad mood was the most central and bridge symptom in the network structure. The other bridge symptoms were irritability and nervousness and stress, which are symptoms of anxiety and perceived stress, respectively. The bridge symptoms increased the risk of developing mental health comorbidities. Therefore, targeting the bridge symptoms can serve as early warning indicators and help improve the early detection of individuals for treatment and prevention of comorbidity.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.

Ethics statements

Patient consent for publication

Ethics approval

This study involves human participants and was approved by the Institutional Review Board of the Addis Continental Institute of Public Health with reference numbers ACIPH/IRB/002/2020 and ACIPH/IRB/005/2021 and the Haramaya University (SPH/01/180135/2021). Verbal informed consent was obtained from all study participants. Only deidentified data were available for this study.

Acknowledgments

The authors acknowledge the study participants and the data collectors for their willingness to participate in this study.

This post was originally published on https://bmjopen.bmj.com