Healthcare utilization during the first two waves of the COVID-19 epidemic in South Africa: A cross-sectional household survey

Healthcare utilization surveys contextualize facility-based surveillance data for burden estimates. We describe healthcare utilization in the catchment areas for sentinel site healthcare facilities during the first year of the COVID-19 pandemic. We conducted a cross-sectional healthcare utilization survey in households in three communities from three provinces (KwaZulu-Natal, Western Cape and North West). Field workers administered structured questionnaires electronically with the household members reporting influenza-like illness (ILI) in the past 30 days or severe respiratory illness (SRI) since March 2020. Multivariable logistic regression was used to identify factors associated with healthcare utilization among individuals that reported illness. From November 2020 through April 2021, we enrolled 5804 households and 23,003 individuals. Any respiratory illness was reported by 1.6% of individuals; 0.7% reported ILI only, 0.8% reported SRI only, and 0.1% reported both ILI and SRI. Any form of medical care was sought by 40.8% (95% CI 32.9% - 49.6%) and 71.3% (95% CI 63.2% - 78.6%) of individuals with ILI and SRI, respectively. On multivariable analysis, respiratory illness was more likely to be medically attended for individuals at the Pietermaritzburg site (aOR 3.2, 95% CI 1.1–9.5, compared to Klerksdorp), that were underweight (aOR 11.5, 95% CI 1.5–90.2, compared to normal weight), with underlying illness (aOR 3.2, 95%CI 1.2–8.5), that experienced severe illness (aOR 4.8, 95% CI 1.6–14.3) and those with symptom duration of ≥10 days (aOR 7.9, 95% CI 2.1–30.2, compared to <5 days). Less than half of ILI episodes and only 71% of SRI episodes were medically attended during the first two COVID-19 waves in South Africa. Facility-based data may underestimate disease burden during the COVID-19 pandemic.


Introduction
The first case of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in South Africa was reported on 5 March 2020.A national lockdown was instituted from 27 March-1 May 2020, which was followed by gradual and phased changing of restrictions in response to the national number of cases [1].By the end of April 2021, South Africa had experienced two waves of infection with the first wave peaking in July 2020, and a second wave peaking in January 2021.
Healthcare utilization surveys (HUS) are useful to characterize the healthcare seeking patterns for diseases of interest.They have been used to complement sentinel surveillance by allowing assessment of the sensitivity of a surveillance system and adjustment of facility-based surveillance data to estimate disease burden within the community [2,3].Understanding healthcare utilization in a community helps to characterize healthcare seeking behaviour, strengthen access to healthcare and improve the interpretation of surveillance data.HUS were conducted in 2012 in Soweto and Klerksdorp communities in South Africa, approximately onethird of individuals with pneumonia did not seek care with a licensed service provider and all individuals with influenza-like illness (ILI) sought care [4].HUS were conducted in Pietermaritzburg in 2013, 4% of individuals with pneumonia and 13% of individuals with ILI did not seek care [5].A cross-sectional study in Johannesburg in 2015 showed that 32% of participants that reported ILI and 0% of participants that reported pneumonia did not seek care [6].However, patterns in healthcare utilization changed during the coronavirus disease 2019 (COVID-19) pandemic, given lockdown restrictions and public fear for the then novel virus [7].
The National Institute for Communicable Diseases (NICD) has conducted syndromic surveillance for severe respiratory illness (SRI) and influenza-like illness (ILI) at sentinel hospitals and primary healthcare clinics in five provinces of South Africa since 2009.In March 2020, the NICD began testing surveillance samples for SARS-CoV-2 to understand the burden of COVID-19.To further evaluate the community burden of disease it is important to adjust the COVID-19 incidence obtained from facility-based surveillance by the healthcare utilization patterns of the sentinel site catchment population.We describe healthcare utilization patterns within the catchment areas of sentinel site healthcare facilities (Edendale Hospital and Edendale Gateway Clinic in Pietermaritzburg, KwaZulu-Natal Province, Mitchell's Plain Hospital and Mitchell's Plain Clinic in Mitchell's Plain, Western Cape Province, and Tshepong Hospital, Klerksdorp Hospital and Jouberton Clinic in Klerksdorp, North West Province).

Study design and population
We conducted a cross-sectional HUS in households in three of the communities serviced by facilities where severe respiratory illness (SRI) and influenza-like illness (ILI) surveillance was conducted [8,9], namely Mitchell's Plain, Pietermaritzburg, and Klerksdorp, using a one-stage cluster sampling design.Mitchell's Plain is a large township in the City of Cape Town, Western Cape Province, Pietermaritzburg is the capital and second largest city in KwaZulu-Natal Province, and Klerksdorp is the largest city in North West Province.Study sites were selected based on having long-term surveillance data available from both primary healthcare clinic (ILI surveillance) and hospital (SRI surveillance) facilities.

Sample size
The sample size was calculated for a one-stage cluster sampling design, with a 95% confidence interval, 10% precision, a 50% expected healthcare seeking among individuals reporting a severe respiratory illness within a predefined period of time in the community and an assumed design effect of 1.5 (household cluster).The sample size was calculated to be 144 individuals reporting a severe respiratory illness in the selected households.Based on data obtained from population-based hospital surveillance and the use of healthcare utilization surveys previously conducted in the target communities, the annual cumulative incidence of severe respiratory illness in these communities is estimated to be 2 per 100 population; therefore, we aimed to interview 7,200 individuals (i.e., 144/0.02) in each community.Assuming an average household size of 3 members, we aimed to enrol 2,400 households in each community.We accounted for a 20% household refusal rate.

Selection and enrolment of households
Households were identified using randomly selected global positioning system (GPS) coordinates.The boundaries of each catchment area were delineated on aerial maps available from Google Earth or the local municipality.Non-residential areas such as parks, industrial areas and sports complexes were excluded.In township areas, we used aerial images of the townships to create polygons corresponding to the townships.Within each township polygon, we randomly sampled geographic coordinates where the number of coordinates was proportional to the population of the township.The household closest (within 30 meters) to each random geographic coordinate was approached for the survey.Fieldworker teams visited each selected household up to three times on separate days or times as needed.A household was excluded if the head of the household/primary caregiver was unavailable after three visits on separate days or times or declined participation in the study.Individuals of all ages that lived in recruited households since 1st March 2020 were eligible to be enrolled in the study, while individuals with unknown residence or residence outside of the catchment area were excluded.Additional GPS co-ordinates were generated at the start of the study and replacement households were visited according to the order on the list.

Data collection
Field workers administered structured questionnaires electronically using Research Electronic Data Capture (REDCap, Vanderbilt University, USA) with the primary caregiver of the household, to gather information on household demographics and screening of household members for symptoms for SRI (sudden onset or worsening fever with cough, and difficulty breathing lasting between 2 and 30 days, or diagnosed with pneumonia) since the beginning of March 2020, or ILI (sudden onset or worsening fever with cough) in the past 30 days.SRI and ILI case definitions were adapted from previous healthcare utilization surveys [2,4,5].After providing written informed consent, household members identified by the primary caregiver as having SRI and/or ILI were interviewed by the fieldworkers.If the participant reported >1 respiratory illness episode, information was collected on the most recent episode of respiratory illness.Demographic information as well as information on underlying illnesses (including tuberculosis (current or previous), asthma, diabetes, chronic heart disease, chronic lung disease, hypertension and cancer), symptoms experienced and healthcare seeking for the reported ILI/SRI episode (clinic, hospital, traditional healer, religious leader, friend/relative, community health worker or pharmacy) were collected from participants.Height and weight were measured to calculate the body mass index (BMI) for individuals aged �5 years.BMI could not be calculated for children aged <5 years as age in month units was not available, and was classified as unknown.Participants were asked if they had been previously tested for SARS-CoV-2 infection and the result thereof.If the household member was aged <18 years, information was obtained from the child's parent/guardian.Fieldworkers also interviewed primary caregivers to collect healthcare seeking information for household members that had died from any cause in the period since beginning of March 2020.

Data analysis
Body mass index (BMI) was calculated using participants measured height and weight, and categorised using WHO standards [10,11].Socioeconomic status (SES) was measured using a standardised set of questions described in previous research [12].Responses to SES questions were summed, a score created and categorised into three levels (low, medium and high).Crowding in the household was defined as a mean of >2 individuals per sleeping room.Continuous variables were summarized using median and interquartile ranges (IQR).Categorical variables were summarized using frequency distributions and compared using Pearson's Chisquared test.Hierarchical multivariable logistic regression, controlling for site and household (within site) clustering, was used to identify factors associated with healthcare utilization among individuals that reported illness, starting with all variables that were significant at p-values <0.2 on univariate analysis and dropping non-significant factors with manual stepwise backward selection.All 2-way interactions were evaluated.Two-sided p-values <0.05 were considered significant.Analysis was performed using Stata 14.1 1 (StataCorp LP, College Station, United States of America).

Ethics
This study was approved by the University of the Witwatersrand (M200861) Human Research Ethics Committee; the U.S. Centers for Disease Control and Prevention Institutional Review Board (7322) relied on the local ethics committee.The study was also approved by the respective community and provincial research committees.The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the CDC.

Discussion
In a household survey conducted in three communities in South Africa after the first two waves of the COVID-19 pandemic (dominated by the ancestral and Beta variants of concern, respectively), 1.6% of individuals reported experiencing respiratory illness.ILI was reported by 0.7% of individuals during the 30 days prior to the interview and severe respiratory illness by 0.8% of individuals since March 2020.The majority of individuals reporting ILI (78.4%) and SRI (91.7%) were older than 18 years.Healthcare utilization was higher for individuals with SRI (71.3%) than those with ILI (40.8%).Respiratory illness was more likely to be medically attended in individuals at the Pietermaritzburg site, underweight individuals, individuals that had underlying illness or severe illness and those with symptom duration �10 days.Among interviewed households, 2.2% reported having had at least one household member die since the start of the COVID-19 pandemic.
In our study in the first year of the COVID-19 pandemic, 0.7% of individuals reported having experienced ILI in the past 30 days and 0.8% experienced SRI since March 2020.During this time, South Africa experienced two epidemic waves dominated by the ancestral virus and Beta variants, respectively.A national lockdown was implemented from 27 March-1 May 2020, followed by gradual and phased changing of restrictions (including non-pharmaceutical interventions such as social distancing and compulsory mask wearing) in response to the national number of cases [1].Previous healthcare utilization surveys have been conducted in sentinel site catchment populations in order to estimate the proportion of illness not captured by routine facility-based surveillance.In a survey conducted in Pietermaritzburg in 2013, 5% of individuals reported ILI in the past 30 days and 0.5% reported pneumonia [5].In a survey conducted in Klerksdorp in 2012, 2% of individuals reported ILI in the past 30 days and 2% reported pneumonia [4].In our study, the proportion of individuals with ILI was lower than in previous studies.This may reflect the observed lower activity of non-SARS-CoV-2 respiratory pathogens due to the enforcement of non-pharmaceutical interventions during the pandemic [13].
During the first year of the COVID-19 pandemic, 59.2% of individuals with ILI and 28.7% with SRI reported did not seek medical care for their illness.In a previous healthcare utilization survey in Pietermaritzburg, 13% of individuals with ILI did not seek medical care [5].In Klerksdorp in 2012, 0% and 28% of individuals with ILI and pneumonia did not seek medical care [4].Compared to previous studies in the same areas, our study showed reduced healthcare utilization during the initial pandemic period.This reflects trends reported in many countries in which there was decrease in healthcare utilization during this period.A systematic review including data from 20 countries reported a 37% reduction in in use of healthcare services [14].A reduction in care seeking was observed in the 18 months following the start of the pandemic in the private healthcare sector in South Africa [15].Similar to our study, the most common reason for not seeking medical care in previous healthcare utilization surveys conducted in South Africa was individuals not feeling sick enough, and the most common facility where care was sought was public clinics [4][5][6].Some individuals also reported not seeking medical care because they were not able to get transport and not feeling that it was safe.
Individuals were more likely to seek medical care if they were underweight, had underlying illness, experienced severe illness or had longer duration of symptoms.On univariate analysis, young (<5 years) and older (�60 years) aged individuals were more likely to seek medical care, although age was not significant in the multivariable model, possibly due to small numbers.In a systematic review of the impact of COVID-19 on healthcare utilization, greater reductions in seeking medical care were observed for milder cases of illness [14].A previous survey in Soweto in 2012, South Africa reported care seeking to have been associated with female sex and age <18 years [4].
Our study had a number of limitations.First, all respiratory illnesses were based on selfreports and a time period 30 days for ILI and approximately one year for SRI which may have been influenced by recall bias.Second, the case definitions used were syndromic and not laboratory-confirmed.These may have resulted in misclassification of ILI and SRI cases.Third, only 67% of households visited were enrolled in the study, and the results may; therefore, not be fully representative of the communities in which the surveys were conducted.Fourth, the number of individuals that did not seek healthcare for their illness is likely underestimated as individuals that had died during the study period were not included.Finally, the number of individuals reporting respiratory symptoms and seeking healthcare during the first two waves of the pandemic were small and we may have been underpowered to detect factors associated with medically attended illness such as age.

Conclusions
Healthcare utilization surveys improve the interpretation of facility-based surveillance data by quantifying the proportion of individuals that did not seek medical care for their illness.In our study, during the first year of the COVID-19 pandemic, only 40.8% and 71.3% of individuals with mild and severe disease respectively, sought medical care for their illness.Healthcare utilization for respiratory illness was severely lower than in surveys done in earlier years.This should be taken into account when quantifying the burden of COVID-19 disease using facility-based data.

Table 2 . Healthcare utilization among individuals reporting respiratory illness by syndrome, Healthcare Utilization and Seroprevalence (HUTS) study, South Africa, November 2020 -April 2021 (N = 316).
Underlying illness includes current/previous tuberculosis, asthma, diabetes, chronic heart disease, chronic lung disease, hypertension and cancer a Reported respiratory illness (SRI only, ILI only or both SRI and ILI) for individuals where information on healthcare seeking was available.bBody Mass Index (BMI) calculated for individuals aged �5 years c