This was a cross-sectional study conducted in the POD of a tertiary hospital in northern Pretoria, Gauteng province, South Africa from 01 June 2022 to 31 August 2022.

The study population was the family caregivers of patients with schizophrenia attending a tertiary hospital in northern Pretoria, South Africa, that provided at least weekly care to the patients. In the study setting, patients with schizophrenia who attend the outpatient psychiatry clinic are usually accompanied by their family caregivers during hospital visits. At other times, these family caregivers visit the clinic unaccompanied to collect chronic medications on behalf of the patients. Our sample was drawn from these family caregivers. All eligible family caregivers were offered the opportunity to participate in the study.

Data were analysed using STATA version 17 (StataCorp, Texas, USA). Categorical variables were expressed as frequency and percentages (Table 1, Table 2, and Table 3). The prevalence of the four categories of severity of caregiver burden (none to low [0–20]; mild to moderate [21–40]; moderate to severe [41–60]; severe [61–88]) was also expressed as frequency and percentage with its 95% confidence interval (CI). The normality of the continuous variables was assessed with multiple measures including the skewness-kurtosis test. Normally distributed data were expressed as mean and standard deviation. Skewed data were expressed as median and interquartile range (IQR). The correlation between continuous variables among sociodemographic characteristics and total caregiver burden score was assessed using Spearman rank correlation. Missing data were excluded from the calculation of p-values.

Simple and multiple linear regression were further conducted with the total burden of care score as the outcome or dependent variable while the sociodemographic characteristics were the explanatory variables. Variables with univariable p < 0.2 were selected for the multivariable linear regression.^35,36 However, the gender of the caregiver, the relationship of the caregiver with the patient, and self-reported income per month were chosen a priori based on theoretical and pragmatic knowledge. The multivariable linear model was built using the backward elimination technique.³⁷ To increase the efficiency of the analysis, the four categories of caregiver burdens were taken as an ordinal or ordered variable³⁸ and a proportional odds model (POM) of ordinal logistic regression was conducted.^39,40,41 Similarly, univariable and multivariable ordinal regression was built as previously described for linear regression. The ologit command in STATA was used, and variables that did not violate the assumptions of POM or rule of parallel odds based on Brant’s test and ‘omodel’ test were utilised for the modelling. Multicollinearity was assessed among the variables using the variance inflation factor and values less than 10 suggest no collinearity.^42,43 Two-tailed test of the hypothesis was assumed, and a p < 0.05 was taken as a statistically significant level. The goodness-of-fit test (Hosmer-Lemeshow, Pulkstenis-Robinson, and Lipsitz chi-square tests) was calculated, and a p > 0.05 showed model fit.

The study protocol was reviewed and approved by the Sefako Makgatho Health Sciences University (SMU) Research and Ethics Committee, reference SMUREC/M/01/2022:PG. Each participant gave informed consent before participating in the study after the nature of the research had been fully explained. The permission to re-use the ZBI-22 was obtained from the publisher, Oxford University Press. To ensure confidentiality, the data collected from participants did not have personal identification of the caregiver such as the name. Rather, a study code was used to de-identify the participants. The collected data were stored in a lockable cupboard with access control.

Table 1 and Table 2 show the sociodemographic characteristics of the caregivers (N = 300), and their mean age was 46.0 ± 14 years. Most caregivers were females (62%), parents (39%), of low-income status (93.7%), had secondary education (70%), resided with the patient (87%), and helped with all troublesome activities (95.3%).

The Cronbach’s alpha of the 22-item ZBI in our study was 0.86. The minimum and maximum total caregiver burden scores were 0 and 75, respectively, while the median was 19.00 (IQR: 13.0–30.8). Among all participants, ZBI items 2, 4–6, 9–14, and 16–19 had a median score of zero out of four. Items 15 and 21 had the highest median scores of 3 (IQR: 2–4) and 3 (IQR: 2–3), respectively. Other items had median scores ranging from 1 to 2. Table 3 shows the prevalence of severity of caregiver burden. The prevalence of different categories of caregiver burden decreased with increasing severity of burden: none to low 55%, mild to moderate 30.67%, moderate to severe 12%, and severe 2.33%.

There was a positive moderate correlation between age (r = 0.4, p < 0.001), duration of caregiving (r = 0.37, p < 0.001) and the total caregiver burden. On the other hand, there was a non-statistically significant negative correlation between the monthly income and total caregiver burden (r = −0.05, p = 0.512).

Table 4 shows that at multivariable linear regression (after adjusting for covariates), the statistically significant explanatory variables were the total duration of months of caregiving, having another family member that requires care, use of alcohol, educational status, marital status, age, residence, and smoking. The R² of these predictors during univariable analysis were: total duration of months of caregiving 0.1232, age 0.1229, marital status 0.1061, having another family member who had needs for care 0.0620, educational status 0.0475, alcohol ingestion 0.0424, place of residence 0.0356, and smoking 0.0309.

In Table 4, the linear regression coefficient (β) shows the expected change in the dependent variable that is caused by one unit change in the predictor variable. For every 1-month increase in the duration of caregiving, the total caregiver burden increased by 5% (adjusted β [aβ] = 0.05, 95% CI: 0.02–0.07, p = 0.002). The caregiver burden score of participants who did not have another family member who requires attention was about 6.6 lesser as compared to participants with another family member who needed care (aβ = −6.62, 95% CI: −11.07 to −2.16, p = 0.004). Furthermore, participants who do not drink alcohol had on average about 6.6 extra burden of care score as compared to those who drank alcohol (aβ = 6.62, 95% CI: −0.49 to 12.01, p = 0.034). The burden of care score generally decreased with increasing level of education. Participants who were 50 years and older had on average about 5.04 extra burden of care score as compared to participants who were younger than 50 years (aβ = 5.04, 95% CI: 0.46 to 9.61, p = 0.031). Participants living outside Gauteng had on average about 3.8 extra burden score compared to participants living in Northern Gauteng. However, there was no statistically significant difference in burden score between participants living in other parts of Gauteng and those living in Northern Gauteng. In addition, a participant who was a divorcee had a higher total burden of care score of 14.72 as compared to the burden among single participants. A habit of not smoking reduces the total burden of caregiving by 6.28 in comparison to smoking. The variance inflation factor of all the covariates was less than 2, suggesting that there was no collinearity among the variables.

After multivariable ordinal logistic regression modelling of the predictors of severe burden of care, the explanatory variables that were statistically significant were having another family member that requires care, gender, age category (with 50 years as threshold), residence, and duration of months of caregiving (Table 5). Notably, reporting results of an odds ratio as a percentage change in the outcome (dependent) variable given an exposure to a risk factor (explanatory variable) is valuable. The formula for the percentage change is odds ratio minus 1 multiplied by 100.⁴⁴ There was a 57% lesser likelihood of having a severe burden of care among participants with no other family member with needs of care as compared to participants who had another family member with needs of care (aOR: 0.43, 95% CI: 0.24–0.78, p = 0.006). Likewise, there were 51% lesser odds of having a severe burden of care among males as compared to females (aOR: 0.49, 95% CI: 0.28–0.87, p = 0.015). The likelihood of having a severe burden of care among older participants who were 50 years and older was about 2.6 times the odds of having a severe burden of care among participants who were younger than 50 years (aOR: 2.55, 95% CI: 1.49–4.36, p = 0.001). The odds of having a severe burden of care increases with increasing distance of the residence from the study site. Thus, the odds of having a severe burden of care were about 76% higher among participants living in other areas of Gauteng as compared to those living in Northern Gauteng (aOR: 1.76, 95% CI: 1.03–2.99, p = 0.038). The likelihood of having a severe burden of care increases by 0.5% for every additional 1 month of caring for the patients (aOR: 1.005, 95% CI: 1.001–1.009, p = 0.006). The self-reported income class of the participants did not influence the risk of having a severe burden of care even after multivariable ordinal regression modelling (POM). There was no violation of the POM based on the result of Brant’s test (X² = 13.71, p = 0.472) and the omodel result (X = 5.57, p = 0.936). The goodness of fit test showed that Hosmer-Lemeshow (p = 0.378), Pulkstenis-Robison (p = 0.981), and Lipsitz (p = 0.063) tests all had p > 0.05.

The median ZBI-22 score was 19.0 (IQR: 13.0–30.5). The explanatory variables that predicted total and severe caregiver burdens in multivariable linear and ordinal logistic regression analyses, respectively, were four: age ≥ 50 years, total duration of months of caregiving, residence farther away from the hospital, and having another family member that needs care.

The median (and IQR) total caregiver burden score was 19.00 (13.00–30.50). Our finding is comparable to the ZBI-22 median score of 11 found by Martyns-Yelllowe⁴⁵ and a mean score of 25.8 ± 7.4 reported by Khalil et al.⁴⁶ However, a mean score as high as 51.7 ± 18.2 and 56.1 ± 17.6 have been reported by Shamsaei et al.² and Yerriah et al.,⁹ respectively. In our study, the distribution of severity of burden among the caregivers was ‘no to low’ 55%, ‘mild to moderate’ 30.7%, ‘moderate to severe’ 12%, and ‘severe’ burden 2.3%. Shamseai et al., in their study, reported that 7.6% of the caregivers had ‘no to low’, 23.5% ‘mild to moderate’, 41.8% ‘moderate to severe’, and 27.1% ‘severe’ burden.² The frequency of burden in our study decreased with increasing severity of the burden category. This is important as only a few participants had severe burden, which may signify that most of the caregivers have good support and coping mechanisms. However, it is difficult to comment further on the possible support and coping mechanisms because of the absence of supporting data as the study’s design did not include data collection on these variables. Although more than half of our participants reported none to low burden, targeted interventions to increase the prevalence of ‘none to low’ burden scores among the caregivers are of essence. This is because a low burden is associated with a high quality of life among caregivers of patients with schizophrenia in South Africa.⁹ It has been reported from South Africa that among family caregivers of patients with schizophrenia and comorbid substance use, little or no burden occurred in 4 (3.9%) while 81 (80.2%) had high or severe caregiver burden.⁹ Differences in patients’ profiles, and caregivers’ personal and caregiving characteristics may account for the variations in the ZBI-22 scores found in various studies. For instance, substance use among patients with schizophrenia is associated with a high caregiver burden.⁹

The four explanatory variables that predicted total and severe caregiver burdens in multivariable linear and ordinal logistic regression analyses, respectively, were age ≥ 50 years, total duration of months of caregiving, residence farther away from the hospital, and having another family member that needs care. Arguably, increasing age and duration of caregiving are associated with increasing responsibility, which may be a source of stress that culminates in burden. For instance, other authors have reported a statistically significant positive correlation between caregiver age and burden.⁴⁷ A counterargument however is that the burden may be reduced if there is skill acquisition and adaptation with increasing age and duration of caregiving. Nonetheless, other authors have shown that a decreased number of contact hours between patients with schizophrenia and their caregivers reduced the caregivers’ distress.⁴⁸ Furthermore, caregivers who live farther away from the hospital are at high risk of travelling more distance to arrive at the hospital to receive services and may incur a high cost of transportation, which may cause financial strain. This is of particular importance in the study setting given that most of the caregivers are of low-income status. Additionally, the multifaceted strains incurred from caring for more than one patient may explain why having another family member who needs care is a determinant of caregiver burden. In previous studies involving family caregivers of patients with schizophrenia, the patient’s age, monthly income, and functioning as well as the caregiver’s gender, level of education, occupation, and the duration of time the caregiver spent with the patient in the same house were predictors of caregiver burden.^49,50 Using the Family Burden Interview Schedule (FBIS), Yu et al. found that gender, education, and having an additional dependent were predictive of caregiver burden in family members of patients with schizophrenia.⁵¹ The variation between the type of predictors identified in our study and those found in other studies may be because of differences in attributes comprising sociodemographic characteristics, strains encountered by the caregivers, and personal perceptions, as these may influence the development and severity of the burden.¹⁵ Furthermore, the type of tool used to measure caregiver burden, and whether or not both total and severe burdens were contemporaneously assessed may explain the variation in findings from different studies.

The R² (coefficient of multiple determination or coefficient of determination) of the multivariable linear regression model was 42.16% (Table 4). This implies that 42.16% of the variance in the dependent variable (total caregiver burden) is explained by the independent variables. Human behaviour is generally difficult to predict and studies about them generally have an R-squared of less than 50%.⁵²

This is one of the few studies on caregiver burden of family caregivers of patients with schizophrenia conducted during the era of coronavirus disease 2019 (COVID-19). To the best of our knowledge, it is the first study that identified the determinants of both total and severe caregiver burdens associated with schizophrenia (contemporaneously in the same group of caregivers). This approach was undertaken to improve the efficiency of predicting both total and severe burdens contemporaneously. Furthermore, we used an instrument that has domains that comprehensively assess caregiver burden including the stressors such as patient’s behaviour as in item 4 in ZBI-22.

The main limitation of our study is that the sociodemographic profile of caregivers may be different in various settings. However, the findings of this study may apply to settings with similar sociodemographic profiles as ours. In addition, the sampling method was a convenience sampling, which may predispose to selection bias. To prevent selection bias, all eligible and consecutively identified caregivers who attended the POD during the period of data collection were included in the study until the sample size was attained. No caregiver declined to participate in the study. Furthermore, there is a lack of robust data on cultural determinants of caregiver burden. Consequently, culturally guided interventions to mitigate the burden of care on family caregivers of patients with schizophrenia require further investigation. This is therefore a future research agenda. Furthermore, we wish to acknowledge that the interpretation and/or expression of the burden may be influenced by the disabilities in the caregivers themselves. Although the chronic illnesses of the participants were ascertained, it is proposed that future studies should address this limitation. Moreover, we did not compare the burden of care across family caregivers of patients with different types of chronic medical conditions. Future studies that will provide this information are required.

A major implication of our study is that it points to recommendations on how to reduce caregiver burden. Firstly, the severity of caregiver burden worsened with increasing duration of months of care. This may suggest that many patients required prolonged long-term care in their activities of daily living because of inadequate or difficult symptom control. Therefore, we recommend regular patient follow-up (as clinically indicated) with easy access to healthcare services. Symptom remission will also allow the caregiver to engage in support groups and skills training/acquisition activities, the latter being crucial since the majority were unemployed.

Secondly, family caregivers aged 50 years and above were identified to be at high risk of having an increased burden; therefore, this age group should be evaluated periodically to determine the severity of their burden. Those with severe burdens should be offered individualised support such as the opportunity for the patients with schizophrenia to be considered for placement in an institution. This also applies to caregivers who have more than one family member that needs care. Of note, accepting placement can be a very difficult decision for many families. Therefore, other alternative interventions should be explored and offered, given that placement shifts some but does not relieve all associated family burdens.⁵³ For every caregiver, however, individualised intervention is crucial because the stressors resulting in the experienced burden may be different. These interventions that may be offered to the caregiver by healthcare professionals to reduce burden include providing education and information, assisting with managing the caregiver’s health conditions, and referring them to other multidisciplinary teams for assistance. Other interventions include to encourage the caregiver to improve self-care and maintain good health, make and execute plans to mitigate identified stressors, use supportive technology where possible, as well as consider respite care services.^53,54

Thirdly, improving transportation services may alleviate the burden among caregivers who reside far away from the hospital. This includes improving both patient-paid public transport systems, and making available efficient ambulance services that will collect patients and caregivers at specific places in the community and convey them to the hospital to receive medical assistance. Currently, in South Africa, patients are transported from their homes to the hospitals only if they have emergency medical conditions and not stable chronic illnesses.