This study used a cross-sectional design to assess the implementation of the COVID-19 home management strategy in Zambia and determine the factors influencing uptake of the model.

The study was conducted in 11 districts purposively selected from the 10 provinces of the country. Zambia is administratively divided into 10 provinces, followed by 116 districts and health facilities. The 11 districts were purposively selected because they reported the highest numbers of confirmed COVID-19 cases during the study period. To enhance representativeness, selection also considered geographic distribution and population diversity to ensure that all 10 provinces of the country were included. However, at the patient level, a systematic random sampling process was applied using the line list provided by each District Health Office (DHO) to select study participants within the chosen districts.¹⁷ This approach enhanced the representativeness and reduced potential selection bias at the individual participant level.

Study participants comprised COVID-19 patients who were managed at home between June 2023 and September 2023 (Figure 1).

The districts were purposively selected because they reported a relatively high burden of COVID-19 cases compared to other districts among the 116 districts of Zambia. In this study, a high number of cases was defined using epidemiological thresholds recommended by the WHO. According to WHO, high-transmission areas are typically defined as those with a test positivity rate greater than 5% or incidence greater than 100 cases per 100 000 population per week.^18,19,20 Similarly, studies in Zambia and the region have used district-level incidence thresholds to classify high-burden areas for surveillance and intervention prioritisation.⁸ The districts represented were from the 10 provinces of Zambia.

A sample size calculation was conducted to determine the number of participants required to compare the proportion of individuals who accepted versus those who did not accept COVID-19 home-based care management. These two groups (Group 1: accepted home-based care; Group 2: did not accept home-based care) were defined based on patients’ reported decisions regarding the home isolation programme.²¹ Based on findings from prior studies and pilot data, it was assumed that approximately 60% of patients advised to isolate at home would accept home-based care, while 40% would not. Using a two-sided Z-test for two proportions, with a significance level (α) of 0.05 and statistical power of 80%, the minimum required sample size was estimated at 98 participants per group (total n = 196) to detect a significant difference between the two proportions (see Equation 1). To account for potential non-response or incomplete data, a 10% buffer was added, resulting in a final target sample size of 108 per group (n = 216 in total). In the absence of prior studies quantifying uptake of home-based care among COVID-19 patients, the sample size was estimated using conventional assumptions for comparing two proportions. A moderate difference of 20% points between groups was assumed (p₁ = 0.60 and p₂ = 0.40). This approach is commonly recommended when empirical data are limited, as it provides a conservative and realistic estimate of the required sample size.^22,23,24

Systematic random sampling was employed to select participants from the list of all COVID-19 patients who were managed under the home-based programme. The sampling frame comprised all patients recorded on the line list across the selected districts during the study period. The total number of eligible patients was first established, after which the sampling interval (k) was calculated by dividing the total number of patients (N) by the desired sample size (n).²⁵ A random starting point between 1 and 3 was generated using a computer-based random number generator, and every kth patient in the ordered list was subsequently selected until the required sample size was reached. The database was exported into Microsoft Excel^®, where patient identification numbers were arranged in ascending order according to the date of enrolment. The use of a computer-assisted approach ensured objectivity and minimised selection bias. In cases where selected records were incomplete or ineligible, the next consecutive record was included to maintain the sampling sequence and achieve the targeted sample size.

The health facility line list from the DHO was used as the sampling frame. Participants were drawn at random from the sampling frame in a systematic manner using a random interval for each district, given the number of eligible participants on the line list. The sampling interval was calculated by dividing the population size by the desired sample size. To ensure a self-weighing sample, a proportion of participants were selected from each district. That is, the number of selected participants varied according to the sampling frame in each area.

To be included in the sample:

The outcome variable uptake (uptake or not) was whether the participant received home-based care or not. Independent variables of interest were age, sex, educational level, employment status, head of home-based care management, whether the patient received encouragement to take home-based care or not, contact with healthcare worker or not, nearest health facility or not and monthly income.

A group of dedicated data collectors conducted interviews in 11 districts. The districts were purposively sampled because they had a high number of positive COVID-19 cases among 116 districts of Zambia. The field team comprised data collectors who were deployed in each district for data collection. These data collectors are fluent in the local languages and understand the culture of the communities. They were trained in research ethics,²⁶ data collection questionnaire implementation and worked under the supervision of a team leader. Data collectors and supervisors were trained on how to complete the questionnaire on the tablet. Data quality was also controlled by close supervision, data cleaning and editing, and cross-checking of the completeness of the questionnaires. The questionnaire was pre-tested in similar settings, which were not part of the study area, and the necessary modifications were made to some items of the questionnaire.

Data from 846 participants were cleaned in Microsoft Excel^®.

Descriptive statistics were employed to summarise participant characteristics. Univariable and multivariable logistic regression were employed to determine the factors that predicted acceptance of home-based care management of patients. An investigator-led multivariable logistic regression was used to select variables that best predicted the outcome, accounting for the weighting. The variables for the final multiple regression model were chosen by first running the multiple logistic regression command with all of the predictor variables, then removing one by one the predictor variables with the highest p-values from the model until only those predictor variables that best predicted the outcome remained in the model. Finally, based on Akaike’s Information Criterion and Bayesian Information Criterion (AIC and BIC) for the competing models, the best-fit model was chosen. The model that had the lowest AIC and BIC values in comparison to other models was picked. The 95% confidence intervals (CI) for the crude odds ratio (cOR) and adjusted odds ratios (aOR) were generated. A p-value less than 0.05 was regarded as significant. R statistical software (version 4.3.2; RStudio, Inc.) was used for all analyses.

Ethical clearance to conduct this study was obtained from the University of Zambia Biomedical Research Ethics Committee (UNZABREC: REF. No. 4452 2023) and the National Health Research Authority (No. NHRA0004/19/10/2023). Anonymity and confidentiality were ensured because the data sets were de-identified; they contained identification numbers rather than participants’ names. All information and survey findings were kept private and securely maintained. Paper records with household identifiers, household and locator information were stored apart from interview records in locked offices and a secure cabinet. The Zambia National Public Health Institute (ZNPHI) was in charge of ensuring that all documents were securely stored.

A total of 566 participants were included in our analysis. Most participants (60%) were female and aged between 35 years and 49 years; 170 (33%), with a median age of 28 years (IQR: 28–50). Nearly half (48%) of the participants had attained secondary education. Additionally, almost two-fifths (43%) of the participants were unemployed.

The majority (88%) of the respondents knew about home management of COVID-19 patients. Of these, 436 (83.7% had received home-based care; n = 85 [16%]) did not receive any home-based care. On monthly income during home-based care, 158 (30%) of the participants earned above 3000.00 Zambian Kwacha (approximately $100.00). The majority (75%) of the participants had sought care from the healthcare worker upon experiencing symptoms of COVID-19. Most (90%) of the participants lived less than 10 km from a health facility (Table 1).

There was a significant difference in acceptance of home-based care management services among patients who were employed, unemployed or self-employed (p = 0.032). Similarly, there was a significant difference (p < 0.001) in the acceptance of home-based care management between patients who had heard about the model and those who had not. Nonetheless, there was no significant difference in acceptance of home-based care management among participants of a particular age group, whether male or female, difference in educational status, monthly income, whether the participants were encouraged to use home-based care or not, whether they went to the health facility or not and whether the participant lived less than 10 km or not from the health facility evidenced by p-value greater than 0.05 (Table 1).

Table 2 presents the final multivariable logistic regression model comparing COVID-19 patients uptake of home-based care management with those who did not. The model demonstrated a good fit to the data. Employed participants had significantly lower odds of uptake of home-based care management compared to unemployed participants, by approximately 62% (AOR = 0.38; 95% CI: 0.18–0.78). Conversely, participants who had previously heard about home management of COVID-19 patients were 6.5 times more likely to uptake home-based care compared to those who had never heard about it (AOR = 6.50; 95% CI: 3.56–11.9). Similarly, those with a monthly income between 600.00 and 1000.00 Zambian Kwacha had 2.6 times higher odds of uptake of home-based care compared to those earning less than 600.00 Kwacha per month (AOR = 2.64; 95% CI: 1.10–6.85). In addition, participants who perceived that the home management model worked well had 4.9 times higher odds of being managed at home compared to those who did not (AOR = 4.90; 95% CI: 2.74–8.68).

Although participants’ education level and income were not statistically significant in the final model, they were retained based on prior empirical evidence showing that these factors are important predictors of acceptance of home-based care. Overall, the comparative analysis highlights that socio-economic factors and prior awareness significantly differentiated patients who accepted home-based care management from those who did not.