COVID-19 hospitalization and mortality and hospitalization-related utilization and expenditure: Analysis of a South African private health insured population.

BACKGROUND: Evidence on the risk factors for COVID-19 hospitalization, mortality, hospital stay and cost of treatment in the African context is limited. This study aims to quantify the impact of known risk factors on these outcomes in a large South African private health insured population. METHODS AND
FINDINGS: This is a cross sectional analytic study based on the analysis of the records of members belonging to health insurances administered by Discovery Health (PTY) Ltd. Demographic data for 188,292 members who tested COVID-19 positive over the period 1 March 2020-28 February 2021 and the hospitalization data for these members up until 30 June 2021 were extracted. Logistic regression models were used for hospitalization and death outcomes, while length of hospital stay and (log) cost per patient were modelled by negative binominal and linear regression models. We accounted for potential differences in the population served and the quality of care within different geographic health regions by including the health district as a random effect. Overall hospitalization and mortality risk was 18.8% and 3.3% respectively. Those aged 65+ years, those with 3 or more comorbidities and males had the highest hospitalization and mortality risks and the longest and costliest hospital stays. Hospitalization and mortality risks were higher in wave 2 than in wave 1. Hospital and mortality risk varied across provinces, even after controlling for important predictors. Hospitalization and mortality risks were the highest for diabetes alone or in combination with hypertension, hypercholesterolemia and ischemic heart disease.
CONCLUSIONS: These findings can assist in developing better risk mitigation and management strategies. It can also allow for better resource allocation and prioritization planning as health systems struggle to meet the increased care demands resulting from the pandemic while having to deal with these in an ever-more resource constrained environment.

Introduction

COVID-19 was declared a Public Health Emergency of International Concern on 30 January 2020, and a pandemic on 11 March 2020 by the World Health Organization (WHO) [1,2]. There is now a body of evidence on the risk factors for COVID-19 related hospitalization and mortality [3-11] from studies in North America, Europe and China. A systematic review and meta-analysis of 102 papers covering 121,437 infected patients found comorbidities such as hypertension, diabetes, cardiovascular diseases, and chronic kidney disease were associated with the severity of COVID-19 infection [8]. Particularly, the elderly and males with underlying diseases were more likely to have severe COVID-19. Of the 102 papers included in the review, 80 were from Asia, 15 from Europe, 11 from North America and 1 from South America but none from Africa. A second systematic review and data synthesis of COVID-19 length of hospital stay across 52 studies, found that patients with COVID-19 in China remained in hospital longer than elsewhere [9]. None of the systematic reviews published to date have included any studies from Africa. The relevance of these studies to the broader African and sub-Saharan context is uncertain given the underlying demographic and disease profile differences between the regions [12].

Evidence on the risk factors in the African or South African context is limited with a review of the literature revealing only three published studies [13-15] which reported on the risk factors for COVID-19 related mortality. Two of these studies involved only public sector patients where HIV and tuberculosis are important risk factors for death. South African studies that have included data from private hospitals have included public vs private sector as a predictor variable rather than exploring outcomes separately in the two sectors (for example by stratifying on health care sector) [16].

Little is known about COVID-19 risk factors in the private sector population or how they may impact on hospital related utilization and expenditure patterns. The for-profit private sector is an important provider of health services in the sub-Saharan African region. A recent WHO report estimates the percentage of health services sourced from private providers in the AFRO region at 40% [17]. The South African health system is highly fragmented with substantial disparities in access, facilities and spending between the government-funded public health system and the private health system. Around 18% of the total South African population is covered by private voluntary health insurance [18] which is the predominant funding mechanism for the private health system. Access to private health services depends mainly on the ability to pay and there are stark racial and socio-economic differences in utilization of public compared to private health services. As of 2018, only 10% of black Africans were members of private medical schemes compared to 73% of white South Africans [19].

It is expected that there are differences in severe disease risk and hospitalization between the public and private sector populations and to date only two studies in South Africa have included data from the private sector [14,20] which found a lower overall case-fatality risk compared to the public sector but did not explore underlying risk factors in the two populations. Evidence on the risk factors for COVID-19 related hospitalization, health care utilization and expenditure patterns are also limited and further evidence is required to confirm and better understand the patterns in this regard.

Relying largely on a fee for service model for provider payment and established clinical coding system, the private voluntary health insurance model generates substantial data enabling analysis of utilization, risk and expenditure of beneficiaries which can provide valuable insights. In South Africa, this approach has been taken to investigate for example, use of antibiotics [21], take-up of influenza vaccines [22] and caesarean section rates [23]. Elsewhere, researchers in the Republic of Korea have used insurance administrative datasets to investigate comorbidities and factors determining medical expenses and length of stay for COVID-19 patients admitted to hospital [24].

The aim of this study was to assess and quantify the impact of known risk factors on COVID-19 hospitalization, hospital related utilization and expenditure, and mortality in a large South African private health insured population over a 12 month period. The results from this study will contribute to addressing the gap in the knowledge base on the actual observed risks and subsequent hospitalization using real world data (RWD). It will enable targeted patient management strategies and risk stratification, identification of opportunities for provider quality and efficiency improvements and will generate information to assess the cost and cost effectiveness of preventative and treatment interventions for patients with COVID-19.

Methods

Study design

This is a cross sectional analytic study based on the analysis of the demographic and claims records of members belonging to medical schemes administered by Discovery Health (PTY) Ltd (DH), one of the largest health insurance administrators in South Africa.

Study population

The study population consisted of 3.5 million individuals from 1.7 million households sharing the same health insurance policy. These policy holders belonged to 19 different health insurance schemes administered by DH, representing around a third of South Africa’s privately insured population. The average family size, average contributions and health care expenditure of the study population were compared to that of the broader health insured population and found to be comparable ensuring that the findings of this study are generalisable to the broader South African population with private health insurance (S1 Table) [25].

Data sources

Secondary de-identified demographic and claims data of members belonging to medical schemes administered by DH for the period from 1 March 2020 to 30 June 2021 was extracted from the data warehouse of Quantium Health, an independent company that provides data analytics and strategic consulting services to DH.

For each insured individual, the data contains the following variables: unique study individual identifier, date of birth, sex and province. For each claim submitted to the administrator for reimbursement of services rendered or items dispensed to an insured individual the data contains the following variables: a unique study individual identifier; dates for the commencement and completion of the service; a code and description for each service rendered/item dispensed, an ICD-10 (10th revision of the International Statistical Classification of Diseases and Related Health Problems) code for the diagnosis of the condition being treated; a Current Procedural Terminology (CPT) code for the procedure carried out; a National Pharmaceutical Product Index (NAPPI) code for any surgical, medical or consumable item dispensed; and the amount being claimed.

Data extraction and classification

The approach used to extract and classify the data is schematically summarized in Fig 1. From all the data for the period, a 3-step approach was used to extract the data. For the first step, individuals who had tested positive for COVID-19 through either the PCR, PKR or real-time RT-PCR tests in the period from 1 March 2020 to 28 February 2021 (study period) were identified and a “demographic extract” consisting of demographic, comorbidity and status elements was extracted for these individuals. For the comorbidity variable, the following conditions were considered as comorbidity risk factors for COVID-19 based on a review of published literature, consultation with the South African-based medical experts overseeing utilisation management at the health insurance, as well as consideration of the health profile of private sector patients in South Africa: Cancer, Chronic Renal Disease, Congestive Cardiac Failure, Chronic Obstructive Pulmonary Disease, Diabetes Mellitus, HIV, Hypercholesterolaemia, Hypertension, Hypothyroidism, Ischaemic Heart Disease, Pregnancy, Tuberculosis. The individuals with these comorbidities were identified using the South African Council for Medical Scheme Guideline algorithms for identifying members with medical conditions using claims records [26].

Fig 1

Data extraction and classification.

For the second step, the claims of the COVID-19 positive individuals over the period from 1 February 2020 to 30 June 2021 were assessed to determine whether they had been hospitalized for the treatment of COVID-19 and a hospital admission indicator was created and added to the demographic extract. Although only individuals who tested positive over the period from 1 March 2020 to 28 February 2021 were included in the study sample, the claims up to 30 June 2021 were included in hospitalization analysis to ensure that the data is not “right censored” as there is lag between testing and hospitalization and between hospitalization and the claims being received by the administrator.

For the third step, for all those COVID-19 positive individuals who had been hospitalized, a “hospital admissions” extract was created consisting of claims, length of stay and treatment marker elements.

To calculate the total cost per hospitalized COVID-19 patient we considered all claims for the hospitalization event including costs for pharmaceuticals, hospital bed charges, consumables, radiology services, general medical practitioner and specialist medical practitioner claims.

Statistical analysis

Four COVID-19 outcomes were analyzed. Two of the outcomes were binary, namely a) whether the patient was hospitalized and b) whether the patient died (here, we assumed all deaths amongst these patients were due to or exacerbated by COVID-19). For the COVID-19 patients who were admitted to a hospital, two further outcomes were analyzed, namely c) length of stay (in days) in the hospital and (d) total cost of claims per patient.

The predictor variables considered in the analysis included age (at time of diagnosis); sex; number of commodities; pandemic wave: (pre-wave 1 (1 March 2020–6 June 2020), wave 1 (7 June 2020–22 August 2020), post wave 1 (23 August 2020–14 November 2020), Wave 2 (15 November 2020 –end Feb 2021)); province: (Eastern Cape, Free State, Gauteng, KwaZulu-Natal, Limpopo, Mpumalanga, North West, Northern Cape, Western Cape); health insurance cover level: (1, 2, 3, 4 where cover level 1 plans offered the lowest level of benefits and cover level 4 plans offered the highest level of benefits) and hospital network: (the six main private hospital networks: A-F). A classification system was used for plans and provider networks to enable blinding of the actual names of plan types and specific hospital providers which is proprietary information. The data are grouped into 19 health regions for the insurance company administrative purposes. In our analyses, the health region was chosen for the random effects to account for potential differences in the population served and the quality of care within different geographic health regions.

Summary statistics included frequencies and percentages for categorical data, and for continuous data median and interquartile range were used. For modelling purposes, two-level random-effects logistic regression models were used for the two binary outcomes, where the level-2 unit was the health region.

Exploratory analysis showed that a Poisson model was insufficient to model the length of hospital stay as the data exhibited overdispersion in the sense that its variance exceeded its mean. Thus, a random effects negative binomial regression model was used for the number of days spent in a hospital and we accounted for health region variation as well as overdispersion. Unadjusted Incidence Rate Ratios (IRR) and adjusted Incidence Rate Ratios (aIRR) are presented.

The total cost data was heavily skewed to the right, and upon taking the logarithm of it, the transformed total cost had a “normal’ shape. Thus, a linear mixed regression model on the log of total cost, again using the health region as a clustering level, was used.

Rather than presenting the estimated coefficients (which are increases in log costs per unit change in the respective predictor variable (category)), the estimated coefficients (e.g. beta1) were expressed as percentage increase or decrease depending on whether the coefficient is positive or negative using the formula, (exp(beta1)-1)x100. The coefficients from the linear regression model are shown in S2 Table.

We performed both univariate and multivariable analyses for all four outcomes (including all the predictors and random effects) to identify independent predictors of the modelled outcomes. The multivariate analyses produced adjusted effects as opposed to unadjusted effects from using univariate analysis.

A further separate analysis was carried out to assess the association between the most common comorbidities and combinations of comorbidities and the of risk of hospitalization and mortality. For this analysis, frequencies and percentages and unadjusted odds ratios (estimated using standard regression models) are reported for two outcome variables–hospitalization and mortality. STATA/SE 16.1 was used for all the analyses.

Ethical considerations

Data for the study was made available as part of Quantium Health’s commitment to support research initiatives with broader public health significance. The company does not advise its clients on the clinical treatment of its members. The data was accessed in terms of and under the conditions set out in the agreement between Quantium Health and Discovery Health and a memorandum of understanding between Quantium Health and the study investigators.

All the data was provided in a de-identified format and aggregated at an individual level and the research team had no access to information that would enable the identification of any individual. All findings are presented at an aggregate level and no confidential member, health care provider or scheme information is disclosed. Ethics approval for the use of the database to carry out this study was granted by the Ethics Committee of the SAMRC (project registration number EC018-4/2021).

Results

Sample description

The total dataset comprised the claims of an average total of 3,48 million individuals over the period 1 March 2020 to 28 February 2021. From this dataset, the claims data related to a total of 188,292 individuals who tested positive for COVID-19 over the period were extracted and analyzed. Of the total cases, 41.4% were aged between 40 and 65 years, 54.3% were female, 37.6% were diagnosed in Wave 1 and 51.1% were diagnosed in Wave 2, 40.9% were from Gauteng province, 65.6% had no comorbidities and 61.3% were on the Cover Level 3 plans (Table 1).

Table 1

Sample description and univariate and multivariate analysis of factors associated with admission to hospital and mortality.

		Hospitalisation			Mortality
Variable	COVID-19 cases n (%)	Hospitalised n (%)	Unadjusted OR (95% CI)	Adjusted OR (95% CI)	Deceased n (%)	Unadjusted OR (95% CI)	Adjusted OR (95% CI)
Age
Less than 18	11,669 (6.2)	1,204 (10.3)	Reference	Reference	14 (0.1)	Reference	Reference
Between 18–25	11,932 (6.3)	621 (5.2)	0.48 (0.43–0.53)	0.48 (0.44–0.53)	13 (0.1)	0.91 (0.43–1.93)	0.95 (0.45–2.03)
Between 25–40	68,196 (36.2)	7,357 (10.8)	1.05 (0.98–1.12)	1.03 (0.96–1.09)	348 (0.5)	4.27 (2.50–7.29)	4.35 (2.55–7.43)
Between 40–65	78,031 (41.4)	17,384 (22.3)	2.49 (2.34–2.65)	1.84 (1.73–1.96)	2,674 (3.4)	29.54 (17.46–49.96)	23.12 (13.65–39.14)
Greater than 65	18,464 (9.8)	8,901 (48.2)	8.09 (7.57–8.64)	4.31 (4.02–4.62)	3,202 (17.3)	174.66 (103.27–295.41)	108.26 (63.85–183.57)
Sex
Female	102,184 (54.3)	17,431 (17.1)	Reference	Reference	2525 (2.5)	Reference	Reference
Male	86,108 (45.7)	18,036 (20.9)	1.29 (1.26–1.32)	1.19 (1.17–1.23)	3726 (4.3)	1.78 (1.69–1.88)	1.61 (1.52–1.69)
Province
Western Cape	37,283 (19.9)	7,248 (19.4)	Reference	Reference	1,183 (3.2)	Reference	Reference
Eastern Cape	11,962 (6.4)	1,902 (15.9)	0.78 (0.74–0.83)	1.06 (0.93–1.20)	461 (3.9)	1.22 (1.09–1.36)	1.48 (1.28–1.72)
Free State	5,730 (3.0)	1,078 (18.8)	0.96 (0.89–1.03)	0.86 (0.77–0.97)	145 (2.5)	0.79 (0.66–0.94)	0.96 (0.77–1.19)
Gauteng	76,249 (40.9)	13,514 (17.7)	0.89 (0.86–0.92)	0.89 (0.82–0.96)	2,278 (3.0)	0.94 (0.87–1.01)	1.26 (1.13–1.39)
KwaZulu-Natal	34,856 (18.7)	7,705 (22.1)	1.17 (1.13–1.22)	1.06 (0.97–1.17)	1,604 (4.6)	1.47 (1.36–1.59)	1.75 (1.58–1.93)
Limpopo	4,091 (2.2)	717 (17.5)	0.88 (0.81–0.96)	1.21 (1.07–1.38)	114 (2.8)	0.87 (0.72–1.06)	1.48 (1.19–1.84)
Mpumalanga	7,096 (3.8)	1,279 (18.0)	0.91 (0.85–0.97)	0.99 (0.90–1.11)	184 (2.6)	0.81 (0.69–0.95)	1.13 (0.95–1.35)
North West	6,184 (3.3)	1,316 (21.3)	1.12 (1.05–1.12)	1.04 (0.93–1.17)	186 (3.0)	0.95 (0.81–1.11)	1.29 (1.08–1.54)
Northern Cape	3,038 (1.6)	599 (19.7)	1.02 (0.93–1.12)	1.14 (1.02–1.29)	79 (2.6)	0.81 (0.65–1.03)	1.36 (1.06–1.74)
# of comorbidities
0	123,507 (65.6)	15,110 (12.2)	Reference	Reference	1,757 (1.4)	Reference	Reference
1	30,668 (16.3)	6,758 (22.0)	2.03 (1.96–2.09)	1.52 (1.47–1.57)	1,100 (3.6)	2.58 (2.39–2.78)	1.29 (1.19–1.40)
2	17,105 (9.1)	5,744 (33.6)	3.63 (3.49–3.76)	2.19 (2.11–2.28)	1,252 (7.3)	5.47 (5.08–5.89)	1.79 (1.65–1.94)
3	10,382 (5.5)	4,243 (40.9)	4.96 (4.75–5.17)	2.64 (2.52–2.77)	1,098 (10.6)	8.19 (7.58–8.86)	2.13 (1.96–2.33)
>3	6,630 (3.5)	3,612 (54.5)	8.58 (8.16–9.04)	3.97 (3.76–4.21)	1,044 (15.7)	12.96 (11.94–14.04)	2.64 (2.41–2.89)
Pandemic wave
Pre-Wave 1	4,882 (2.59)	1,091 (22.3)	1.42 (1.33–1.53)	1.49 (1.38–1.61)	186 (3.8)	1.35 (1.15–1.57)	1.39 (1.18–1.64)
Wave 1	70,742 (37.57)	11,884 (16.8)	Reference	Reference	2,021 (2.9)	Reference	Reference
Post-wave 1	16,475 (8.75)	3,662 (22.2)	1.41 (1.36–1.47)	1.47 (1.41–1.54)	386 (2.3)	0.81 (0.73–0.91)	0.72 (0.64–0.81)
Wave 2	96,193 (51.09)	18,380 (19.6)	1.20 (1.17–1.24)	1.18 (1.15–1.21)	3,658 (3.8)	1.34 (1.27–1.42)	1.21 (1.14–1.28)
Medical insurance cover level
Level 1	26,512 (14.3)	4,796 (18.1)	Reference		856 (3.2)	Reference
Level 2	22,460 (12.1)	4,554 (20.3)	1.15 (1.10–1.20)		804 (3.6)	1.11 (1.01–1.23)
Level 3	113,721 (61.3)	21,088 (18.5)	1.03 (0.99–1.07)		3,531 (3.1)	0.96 (0.89–1.04)
Level 4	22,835 (12.3)	4833 (21.2)	1.22 (1.16–1.27)		1,019 (4.5)	1.39 (1.28–1.53)
Total	188,292	35,467 (18.84)			6,251 (3.32)

OR = odds ratio.

Hospitalization risk

The overall hospitalization risk for COVID-19 positive individuals was 18.8% (Table 1). Age, sex and number of comorbidities were found to be independent predictors of hospital admission. Patients aged above 65 years (aOR 4.31; 95%CI 4.02–4.62); who were males (aOR 1.19; 95%CI 1.17–1.23) and had more than three comorbidities (aOR 3.97; 95% CI 3.76–4.21) were more likely to be admitted to hospital. Pre-wave 1 period (aOR 1.49; 95%CI 1.38–1.61), post-wave 1 (aOR 1.47; 95%CI 1.41–1.54), and wave 2 (aOR 1.18; 95%CI 1.15–1.21) all had a higher hospitalization risk compared to wave 1. Provincial differences in hospitalization risk were also observed with admissions more likely in Limpopo and the Northern Cape and less likely in Free State and Gauteng, compared with the Western Cape province. Health insurance cover level 4 (the most expensive plan with the highest level of insurance cover) was associated with a higher risk of hospitalization compared to plan level 1 (OR 1.22; 95% CI 1.16–1.27) in univariate analyses (Table 1). However, we did not include health insurance in the multivariable analyses because it was highly correlated with both age and number of comorbidities, which could have resulted in multicollinearity problems. Sixty-seven percent of individuals on plan level 4 were over the age of 40 and of those with more than 3 comorbidities almost a quarter (23%) were on plan level 4, compared with only 14% of those with 1 comorbidity.

Mortality risk

The overall mortality risk for COVID-19 positive individuals was 3.3% (Table 1). In multivariable analysis, after adjustment for all other factors, mortality risk was the greatest for those aged above 65 year (aOR 108.26; 95% CI 63.85–183.57); males (aOR 1.61; 95% CI 1.52–1.69) and those with more than three comorbidities (aOR 2.64; 95% CI 2.41–2.89). Mortality risk was higher in pre-wave 1 (aOR 1.39; 95% CI 1.18–1.64) or Wave 2 (aOR 1.21; 95% CI 1.14–1.28) compared to wave 1. In terms of provincial differences, all provinces except Free State and Mpumalanga had a significantly higher risk of mortality compared to the Western Cape with the highest being KwaZulu-Natal Province (aOR 1.75; 95% CI 1.58–1.93) (Table 1). In univariate analysis medical insurance cover level 4 was associated with a higher risk of mortality compared to plan level 1 (OR 1.39; 95% CI 1.28–1.53) (Table 1).

Hospitalization utilization

The overall median length of hospital stay for COVID-19 positive individuals was 6 days (IQR 3–10) (Table 2). In multivariable analysis, there was an increasing trend in length of hospital stay with age. Those aged over 65 years had a two-fold increased length of hospital stay compared with those less than 18 years (aIRR 2.00; 95% CI 1.89–2.12). Males had longer hospital stays than females (aIRR 1.08; 95% CI 1.06–1.09) and an increase in length of stay was observed for each additional comorbidity with individuals experiencing more than three comorbidities having the longest period of hospitalization (aIRR 1.14; 95% CI 1.11–1.18). Length of hospital stay was longer in wave 2 compared to wave 1 (aIRR 1.03; 95% CI 1.01–1.05). Provincial differences in length of hospitalization were observed with Eastern Cape, Free State, Gauteng, KwaZulu-Natal and North West all having significantly longer median length of hospitalization of COVID-19 patients compared to the Western Cape (Table 2).

Table 2

Univariate and multivariate analysis of factors associated with hospitalisation length of stay.

Hospital utilisation (days)
Variable	Total hospitalised COVID-19 cases	Median length of hospitalisation (IQR)	Unadjusted length of hospitalisation IRR (95% CI)	Adjusted length of hospitalisation aIRR (95% CI)
Age
Less than 18	1,204	3 (1–4)	Reference	Reference
Between 18–25	621	3 (2–5)	1.13 (1.03–1.23)	1.09 (1.00–1.19)
Between 25–40	7,357	4 (2–7)	1.34 (1.26–1.41)	1.36 (1.28–1.44)
Between 40–65	17,384	7 (4–11)	2.12 (2.01–2.24)	1.80 (1.71–1.90)
Greater than 65	8,901	8 (4–14)	2.54 (2.41–2.69)	2.00 (1.89–2.12)
Sex
Female	17,431	5 (3–10)	Reference	Reference
Male	18,036	6 (4–11)	1.17 (1.14–1.19)	1.08 (1.06–1.09)
Province
Western Cape	7,248	5 (3–10)	Reference	Reference
Eastern Cape	1,902	6 (3–11)	1.05 (1.01–1.10)	1.09 (1.03–1.15)
Free State	1,078	6 (3–10)	1.03 (0.97–1.01)	1.08 (1.01–1.15)
Gauteng	13,514	6 (3–11)	1.05 (1.03–1.08)	1.09 (1.06–1.14)
KwaZulu-Natal	7,705	7 (4–11)	1.08 (1.05–1.11)	1.14 (1.09–1.18)
Limpopo	717	5 (3–8)	0.87 (0.81–0.93)	0.97 (0.90–1.05)
Mpumalanga	1,279	5 (3–10)	0.99 (0.94–1.04)	1.02 (0.97–1.08)
North West	1,316	7 (3–11)	1.06 (1.00–1.11)	1.11 (1.05–1.18)
Northern Cape	599	5 (3–8)	0.89 (0.82–0.96)	0.99 (0.92–1.06)
# of comorbidities
0	15,110	5 (3–9)	Reference	Reference
1	6,758	6 (3–11)	1.25 (1.22–1.28)	1.08 (1.06–1.11)
2	5,744	7 (4–12)	1.39 (1.36–1.43)	1.13 (1.10–1.16)
3	4,243	7 (4–13)	1.44 (1.39–1.48)	1.11 (1.08–1.14)
>3	3,612	8 (4–13)	1.55 (1.50–1.60)	1.14 (1.11–1.18)
Pandemic wave
Pre-Wave 1	1,091	6 (3–12)	1.09 (1.03–1.14)	1.06 (1.01–1.11)
Wave 1	11,884	6 (3–11)	Reference	Reference
Post-wave 1	3,662	5 (2–9)	0.88 (0.85–0.90)	0.89 (0.87–0.93)
Wave 2	18,380	6 (3–11)	1.31 (1.27–1.35)	1.03 (1.01–1.05)
Medical insurance cover level
Level 1	4,796	6 (3–10)	Reference
Level 2	4,554	6 (3–10)	1.00 (0.96–1.04)
Level 3	21,088	6 (3–10)	1.00 (0.97–1.03)
Level 4	4,833	7 (3–12)	1.11 (1.07–1.15)
Private Hospital network
Network A	1,307	6 (4–10)	Reference
Network B	9,601	7 (4–11)	1.09 (1.03–1.15)
Network C	546	5 (3–9)	0.77 (0.70–0.84)
Network D	5,407	6 (3–11)	1.04 (0.98–1.09)
Network E	9,151	5 (3–10)	0.93 (0.88–0.98)
Network F	8,846	6 (3–11)	1.02 (0.97–1.07)
Total	35,467	6 (3–10)

aIRR = adjusted incidence risk ratio.

In the unadjusted model, assessing the effect of insurance cover, only insurance plan level 4 (the most expensive plan with the highest level of cover), was significantly associated with length of hospital stay but this effect was not significant in the adjusted model due to the variable being highly correlated with age and number of comorbidities. In univariate analysis, hospital network B had significantly longer hospital stays compared with network A hospitals, but the effect was not significant in the multivariable model.

Hospitalization expenditure

The overall median hospitalization cost per COVID-19 positive case was R49,836 (IQR R28,464—R107,020) (Table 3). After adjustment for all other factors, there was an increasing hospitalization cost with each age category and those over age 65 years incurred a 172% increased cost of hospitalization compared to individuals under age 18 years (95% CI 153.45% - 191.54%). The cost of hospitalization for males was 18% higher than that for females (95% CI 16.18%– 20.92%). Cost of hospitalization increased for each additional comorbidity. Those with more than three comorbidities had 28% higher hospitalization costs compared with individuals with no comorbidities (95% CI 23.37%– 33.64%). With regard to pandemic wave period, hospitalization during wave 2 was 7% more costly compared to the wave 1 period (95% CI 4.08% - 9.42%). With regard to provincial differences, Gauteng and KwaZulu-Natal were both significantly more costly than the Western Cape (11% and 4% more costly respectively), whilst Limpopo and the Northern Cape were less costly compared to the Western Cape (14% and 10% less costly respectively). Hospitalization cost differences were noted between insurance plan levels, with plans 2, 3 and 4 being 8% more costly compared with level 1 plans We also observed differences in cost across hospital networks after adjusting for all other covariates. Hospital networks C, D, E and F were all significantly less costly compared to network A (27%, 12%, 17% and 12% less costly respectively) (Table 3).

Table 3

Univariate and multivariate analysis of factors associated with hospitalisation cost.

Cost per hospitalised COVID-19 patient (SA Rands)
Variable	Total hospitalised COVID-19 cases	Median cost/ patient (IQR)	Unadjusted % Difference (Increase/decrease in cost) (95%CI)	Adjusted % Difference (Increase/decrease in cost) (95%CI)
Age
Less than 18	1,204	24,400 (15,552–39,586)	Reference	Reference
Between 18–25	621	30,390 (18,944–54,741)	20.92 (9.42, 33.64)	27.12 (13.88, 40.49)
Between 25–40	7,357	42,051 (24,974–68,435)	61.61 (52.20, 73.33)	60.00 (50.68, 71.60)
Between 40–65	17,384	57,801 (33,276–138,005)	158.57 (143.51, 174.56)	131.64 (118.15, 145.96)
Greater than 65	8,901	83,335 (41,290–203,098)	215.82 (9.42, 235.35)	171.83 (153.45, 191.54)
Sex
Female	17,431	50,162 (28,650–103,434)	Reference	Reference
Male	18,036	60,793 (33,151–155,153)	27.12 (23.37, 29.69)	18.53 (16.18, 20.92)
Province
Western Cape	7,248	47,080 (26,290–103,145)	Reference	Reference
Eastern Cape	1,902	48,843 (28,421–101,642)	1.01 (-3.92, 7.25)	-1.00 (-5.82, 5.13)
Free State	1,078	45,547 (27,073–84,580)	-5.82 (-12.19, 1.01)	-1.98 (-7.69, 5.13)
Gauteng	13,514	54,059 (30,663–117,780)	15.03 (11.63, 18.53)	11.63 (8.33, 15.03)
KwaZulu-Natal	7,705	50,388 (29,753–107,514)	9.42 (5.13, 12.75)	4.08 (1.01, 8.33)
Limpopo	717	36,514 (21,760–73,986)	-20.55 (-26.66, -13.06)	-14.79 (-21.34, -7.69)
Mpumalanga	1,279	41,264 (23,406–91,366)	-10.42 (-16.47, -4.88)	-3.92 (-10.42, 2.02)
North West	1,316	47,634 (27,784–88,583)	-1.00 (-7.69, 5.13)	-0.10 (-5.82, 6.18)
Northern Cape	599	39,213 (23,116–69,723)	-19.75 (-26.65, -12.19)	-10.42 (-17.30, -1.98)
# of comorbidities
0	15,110	42,079 (24,547–76,712)	Reference	Reference
1	6,758	50,359 (29,152–107,048)	24.61 (20.92, 28.40)	7.25 (4.08, 11.63)
2	5,744	58,489 (33,018–132,798)	46.23 (40.49, 50.68)	17.35 (13.88, 20.92)
3	4,243	62,213 (34,300–148,067)	55.27 (50.68, 61.61)	19.72 (16.18, 24.61)
>3	3,612	73,423 (37,685–170,703)	73.33 (66.53, 80.40)	28.40 (23.37, 33.64)
Pandemic wave
Pre-Wave 1	1,091	52,011 (27,318–120,072)	6.18 (-0.40, 13.88)	13.88 (6.18, 20.92)
Wave 1	11,884	48,577 (27,537–103,472)	Reference	Reference
Post-wave 1	3,662	45,169 (25,009–90,903)	-10.42 (-13.93, -6.76)	-2.96 (-6.76, 1.01)
Wave 2	18,380	51,524 (29,882–112,648)	7.25 (5.13, 10.52)	7.25 (4.08, 9.42)
Medical insurance cover level
Level 1	4,796	44,071 (24,695–91,445)	Reference	Reference
Level 2	4,554	50,698 (29,034–114,502)	20.92 (15.03, 25.86)	8.65 (4.08, 13.43)
Level 3	21,088	49,326 (28,477–102,979)	16.18 (12.75, 19.72)	8.76 (5.13, 12.75)
Level 4	4,833	59,609 (32,486–131,414)	37.71 (32.31, 43.33)	8.65 (4.08, 13.43)
Private Hospital network
Network A	1,307	53,407 (30,403–123,034)	Reference	Reference
Network B	9,601	59,086 (35,733–128,490)	11.63 (5.13, 18.53)	5.13 (-1.00, 11.63)
Network C	546	36,068 (22,360–69,732)	-35.60 (-42.31, -28.11)	-27.39 (-34.30, -18.94)
Network D	5,407	49,165 (27,202–101,265)	-12.19 (-17.30, -5.82)	-12.19 (-17.30, -6.76)
Network E	9,151	44,706 (25,529–91,528)	-17.30 (-22.12, -12.19)	-17.30 (-22.12, -12.19)
Network F	8,846	47,670 (27,554–105,322)	-10.42 (-15.63, -3.92)	-12.19 (-17.30, -6.76)
Total	35,467	49,836 (28,464–107,020)

Risk by comorbidity condition type

Of the conditions considered as comorbidity factors, Diabetes Mellitus (on its own or in combination with other comorbidities) carried the highest hospitalization risk (OR 3.6; 95% CI 3.27–3.94 for Diabetes Mellitus only; OR 6.6; 95%CI 5.88–7.43 for Diabetes with hypertension, hypercholesterolemia and Ischemic heart disease) (Table 4). In terms of mortality risk, the combination of diabetes with hypertension, hypercholesterolemia and Ischemic heart disease carried the highest mortality risk (OR 10.25; 95% CI 8.57–12.27). Hypertension in combination with heart disease (OR 6.94; 95% CI 5.66–8.51) or cancer (OR 6.10; 95% CI 4.47–8.33) also carried an increased risk for mortality (Table 4).

Table 4

Hospital risk and mortality risk by co-morbidity condition combinations.

Condition/Condition combinations	Total cases	Hospitalised n (%)	Unadjusted OR (95% CI)	Deaths n(%)	Unadjusted OR
No Co-morbidities	123,507	15,110 (12.2)	1.0(0.98–1.02)	1,757 (1.4)	1.00(0.94–1.07)
Other co-morbid combinations	16,018	7,034 (43.9)	5.6 (5.42–5.82)	1,844 (11.5)	9.01 (8.42–9.64)
DM-HT-HC-IHD	1,141	547 (47.9)	6.6 (5.88–7.43)	147 (12.9)	10.25 (8.57–12.27)
DM-HT-HC	4,578	1,876 (41.0)	5.0 (4.68–5.29)	449 (9.8%)	7.54 (6.77–8.4)
DM-HT	2,601	1,061(40.8)	4.9 (4.56–5.35)	287 (11.0)	8.59 (7.53–9.8)
HT-HC-IHD	1,176	428 (36.4)	4.1 (3.64–4.62)	107 (9.1)	6.94 (5.66–8.51)
DM-HC	1,484	532 (35.8)	4.0 (3.6–4.46)	85 (5.7)	4.21 (3.37–5.27)
HT-CA	544	187 (34.4)	3.8 (3.15–4.49)	44 (8.1)	6.10 (4.47–8.33)
DM	2,085	695 (33.3)	3.6 (3.27–3.94)	102 (4.9)	3.56 (2.9–4.37)
CA	1,034	325 (31.4)	3.3 (2.88–3.76)	61 (5.9)	4.34 (3.34–5.65)
HIV-HT	755	22 (9.7)	3.0 (2.59–3.55)	44 (5.8)	4.29 (3.15–5.84)
HT-HC	3,849	1,133 (29.4)	3.0 (2.78–3.21)	268 (7.0)	5.19 (4.54–5.93)
HT-HTH	1,067	297 (27.8)	2.8 (2.42–3.17)	67 (6.3)	4.64 (3.61–5.97)
HT	14,167	3,609 (25.5)	2.5 (2.35–2.55)	703 (5.0)	3.62 (3.31–3.96)
HC	2,201	442 (20.1)	1.8 (1.62–2)	56 (2.5)	1.81 (1.38–2.37)
HIV	4,541	806 (17.7)	1.6 (1.43–1.68)	93 (2.0)	1.45 (1.17–1.79)
COPD	6,009	941 (15.7)	1.3 (1.24–1.43)	107 (1.8)	1.26 (1.03–1.53)
HTH	1,537	220 (14.3)	1.2 (1.04–1.39)	30 (2.0)	1.38 (0.96–1.99)

Discussion

This is the first study describing risk factors for COVID-19 hospitalization and mortality and hospitalization related utilization and expenditure amongst a private health insured population in Africa. From a study population of 188,292 COVID-19 cases, we found overall hospitalization rates and mortality rates of 18.8% and 3.3% respectively.

COVID-19 positive individuals above the age of 65 years, those with 3 or more comorbidities and males had the highest risk across all 4 outcome measures. Overall, in line with studies carried out elsewhere [24], the findings suggest that the strongest predictors for COVID-19 related hospitalization, mortality [10,11], hospital related utilization [27] and expenditure [28] was age, followed by the number of comorbidities and then sex.

Regarding specific comorbidities, diabetes alone or in combination with hypertension, hypercholesterolemia and ischemic heart disease carried the greatest risk for hospitalization and death. These comorbidity risk factors for severe disease and death are similar to other settings. In contrast to research amongst public sector COVID-19 patients in South Africa [13], we did not find an association between HIV and mortality reflecting the different underlying disease profile of the private sector population. Around 4.7% of the private insured population in South Africa are registered on an HIV management program [25] whilst the HIV prevalence rate in the general population is 14% [29]. Another recent study in South Africa exploring risk factors for COVID-19 related in-hospital mortality found an HIV prevalence amongst hospitalized Covid-19 patients of 20.4% in the public sector and 2.2% in the private sector [16].

Research based on data from the national surveillance system, including both public and private sector patients, has reported a case fatality risk of 18.7% amongst hospitalized COVID-19 patients in the private sector and 27.5% amongst public sector patients [14]. Our mortality risk does include some deaths (509) amongst individuals who were never hospitalized although almost all (92%) of the deaths in our sample occurred in hospital. Amongst hospitalized cases in our study the mortality risk is 16% (5742/35467) which compares well to the rate reported from national surveillance amongst private sector COVID-19 admissions. Differences in mortality risk between the public and private sector are expected due to differences in underlying disease profiles of patients, resourcing and case load differences.

Provincial variation in all four outcome measures were found, even after adjustment for all other factors. This reflects differences in clinical practice between private hospitals which may not necessarily adhere to national Department of Health COVID-19 protocols. There may also be underlying differences in health seeking behaviour across provinces and different thresholds applied by general practitioners regarding when to admit patients to hospital. With regard to differences in hospitalization cost, Gauteng was almost 12% more expensive compared to the Western Cape. This is likely due to the reduced plan costs for coastal versus inland hospitals under the DH plan options [30].

We found higher rates of hospitalization and mortality and longer duration of hospital stay during wave 2 compared with wave 1. This is similar to findings from the national surveillance system and is likely due to the higher incidence of COVID-19, greater pressure on hospitals and the emergence of the Beta variant [14].

In unadjusted analysis, individuals on level 4 insurance plans were found to have an increased risk of both hospitalization and mortality. This was related to individuals on these plans being older and with more comorbidities reflecting the trend for individuals to buy more expensive and comprehensive medical insurance as they get older and sicker [31].

Hospital networks also differed in the cost of COVID-19 care after adjustment for risk factors. While this could be due to differences in the reimbursement rates across the various hospital network groups, variation in the underlying clinical management of patients across hospital groups is likely to have played a role.

This analysis has a number of limitations. Only services for which claims were submitted were analysed. This could result in under-recording of the COVID-19 cases, particularly of “milder” cases and could result in an overstatement of the reported outcome measures. As this data set is from a period prior to the mass distribution of home-based COVID-19 self-test kits, COVID testing was by doctor referral available under private insurance at no charge where the patient tested positive. For those paying out of pocket the test was relatively expensive (US$55 per test). It is possible that some patients with private voluntary insurance elected to seek testing in the public system or pay out of pocket for the test in which case they would be missed from the data set. However, we consider that this number would be very small as there was strong incentive for those insured to utilize their insurance benefits, and the requirement for a doctor referral also strengthened capture of data. We further note that (1) our study population was limited to those who tested COVID-19 positive and (2) all the statistics that we present are based on those who tested positive (not the entire insured population).

In addition, the costs do not reflect additional out of pocket expenditure for claims not covered under the benefits of certain health plans, for example the use of the pharmaceutical Remdesivir as an adjunct to existing treatments covered by the insurance plan.

Obesity and smoking, identified as risk factors in other studies, could not be included in our analysis due to limited or incomplete coding of these risks in our dataset. We have however included the co-morbidity chronic obstructive pulmonary disease and other large analyses that have included both smoking status and chronic pulmonary disease in adjusted models have noted a mediating effect which limits the ability to explore the independent association of smoking status [32]. Vaccination status could also not be included due to the timeframe for the analysis in relation to the vaccine roll-out. Vaccination roll-out in South Africa began in May 2021 with individuals over the age of 60 years and our analysis included hospitalization data up until 20 June 2021.

The potential for overcrowding in health facilities leading to increased death due to COVID-19 represented a significant concern for South Africa’s COVID-19 response. Differentiating risk by pandemic wave period allowed this analysis to partially account for impact due to overcrowding, however it is unknown whether overcrowding had a differentiated impact on COVID-19 outcomes contingent upon particular risk factors.

This analysis has significant policy implications for both private and public sectors in South Africa for targeted, risk-based interventions and for reducing unwarranted variation. The outputs from the study can provide a basis for developing “risk calculators” to enable providers and funders to develop risk-based management strategies. Risk information can also inform broader policy, including risk stratification for employer “work from home” policies and other initiatives to reduce risk of infection.

The finding of provincial and hospital group variation on outcomes after adjusting for other risk factors is in line with the findings of the Competition Commission’s Health Market Inquiry which identified this variation as a major source of private sector inefficiency in South Africa [33]. The results highlight the difficulties related to efforts to contain health system costs, with complex dynamics between independent clinician judgment, hospital groups, and insurance plan types in a system with few mechanisms for standardization, and even built-in efficiency impediments–for example health insurance providers are by law required to negotiate separately with individual hospital groups. They also highlight the need to identify and minimize unwarranted variation through the implementation of protocols which are evidence based, effective and cost-effective across the private sector. In addition, the analysis provides a basis for determining the cost effectiveness of different treatment and preventative interventions–an understanding of the expected cost and mortality risk in the South African context for different patient populations following a COVID-19 diagnosis enables realistic estimations about how best limited resources can be used in developing clinical guidelines and protocols.

Finally, this analysis demonstrates the untapped potential of RWD in health policy decision making and planning. Through the use of relatively simple multiple regression analysis of a substantive data set, insights were achievable in terms of variation on a range of outcomes, much of which would be unachievable in even a large-scale clinical trial. Using more extensive data analysis and time series, this dataset may also enable testing of various interventions and disease management dynamics, most notably the impact of vaccination and other treatment strategies. This analysis was conducted with ethics approval and maintained confidentiality of individual patient data which demonstrates a workable model of analysis. The larger lesson for health systems is that data systems across public and private sectors must be improved to enable use of data and analytics in decision making. Traditional analytical approaches for informing health investment and planning decisions rely on modeled prevalence and epidemiological information with intervention effects from published literature. While this approach enables evidence-informed decision making, it will always be limited to the extent that it reflects the actual context of the health system, inefficiencies and variations. The advancing digitization of health systems means that the routine use of health system information for investment, planning, efficiency and quality improvement is possible if the appropriate structures are present.

Conclusion

The information from this study, with one of the largest private sector patient datasets, can assist in developing better risk mitigation and management strategies. It can also allow for better resource allocation planning and prioritization strategies as health care systems struggle to meet the immediate and longer term increased health care demands resulting from the COVID-19 pandemic while having to deal with these in an ever-more resource constrained environment [34].

Comparison of Discovery Health administration profile versus rest of the rest of the insured population in South Africa.

(DOCX)

Click here for additional data file.

Univariate and multivariate analysis of factors associated with hospitalisation cost with coefficients.

(DOCX)

Click here for additional data file.

Transfer Alert

This paper was transferred from another journal. As a result, its full editorial history (including decision letters, peer reviews and author responses) may not be present.

22 Feb 2022

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Raymond Nienchen Kuo, Ph.D

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Please include your tables as part of your main manuscript and remove the individual files. Please note that supplementary tables (should remain/ be uploaded) as separate "supporting information" files.

3. We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: I Don't Know

Reviewer #2: I Don't Know

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors present a study looking at the risk factors for COVID-19 hospitalization, mortality, hospital stay and cost of treatment in South Africa.

I think this is an important study which adds a lot of value, particularly as it covers a region of the world for which there has been limited information. The introduction is well written and interesting. It clearly identifies the research question and highlights the importance of this study. It is a rich dataset, which covers a long period of time, not just the beginning of the pandemic, allowing for changes in hospitalisation, mortality, hospital length and cost to be observed. However, I do have several concerns about the methodology which I have outlined below. Unfortunately, Tables 1-4 also seem to be missing from the manuscript, without which it makes it very difficult to assess the results. I have done my best without them, but would appreciate if a copy of the tables could be sent to me so that I can review the results properly. I also think there lacks interpretation and contextualisation of the results in the discussion (highlighted below).

I think if the authors could address these concerns methodological concerns, and expand the interpretation in the discussion, it is a very worthwhile paper.

1. In your population, do you know all people who have received a positive test? Or is it possible that individuals could have received a positive test from a different source, and this is not recorded in the data? This is an important point and should be made clear in the paper. If it is possible that positive tests could have been missed, then this warrants further discussion (i.e. if those individuals did not need hospitalization, this would over-estimate the estimates of hospitalisation and mortality).

2. I think there are comorbidities that are known risk factors for COVID-19 which have not been included in the study.

You state in the methods:

“Conditions were considered as comorbidity risk factors for COVID-19 based on a review of published literature: Cancer, Chronic Renal Disease, Congestive Cardiac Failure, Chronic Obstructive Pulmonary Disease, Diabetes Mellitus, HIV, Hypercholesterolaemia, Hypertension, Hypothyroidism, Ischaemic Heart Disease, Pregnancy, Tuberculosis”

There are conditions that are considered to be risk factors for COVID-19 hospitalisation and death which are not included in this list, i.e. Asthma, chronic liver disease, neurological diseases (stroke/dementia/etc), organ transplant, immunosuppressive condition (rheumatoid arthritis/lupus/etc), in addition to others. For example, see this paper: https://www.nature.com/articles/s41586-020-2521-4#Sec2

3. There are also risk factors such as smoking and obesity that have not been included presumably due to data limitations. It is mentioned in the discussion that obesity could not be included, but I think further discussion on impact of not including them and how this may have affected the results is warranted. (i.e. discussion of the smoking rates/obesity rates)

4. Table 1 - 4 seem to be missing from the manuscript. I have done my best to assess the results without them, but it would be really helpful to see these tables.

5. It is also not completely clear why vaccination could not be included within this analysis. It would be useful to have more of a discussion on the effect of vaccination on hospitalisations and length of stay, and more information on vaccine roll out in South Africa (i.e. % of people vaccinated with 1 or 2 doses during the different waves), and how this would have effected the results observed.

6. I don’t think this is included in the Tables, but I would be interested to see a summary of the different comorbidities (i.e. XX reported Diabetes, XX Cancer; rather than just 0, 1 comorbidities). Perhaps this could be a summary table of the study characteristics. I appreciate this might be a lot of different risk factors and you may have small numbers for some, so perhaps the main comorbidities rather than 0,1,2,>3. Apologies if this is already included within the table 1.

7. Adjusted odds ratios are reported, but it is not clear from the methods exactly what you have adjusted by. This is also true for the reported adusted IRR. I think this could be clarified in the methods (i.e. in the results you say that you did not adjust by medical insurance level for mortality risk as this was highly correlated with age and comorbidities, was this also the case for the hospitalization risk and length of stay/utilization?)

8. I think one of the key strengths of this study is the longer time period over which data was collected, yet this is not really touched on in the discussion. It is not clear to me why mortality and hospitalization would be lower in wave 1? I would have thought that this was before vaccination, and so mortality might have been higher? I think some further discussion would be useful to help readers contextualise these results.

Some minor points:

1. This sentence about the study size in the methods is a little confusing:

“The study population consisted of families (1.7 million) and individuals (3.5 million) belonging to 19 health insurances administered by DH, representing around a third of South Africa’s privately insured

population.”

It is confusing as to whether it is individuals + families, or total individuals, and of these, there are 2.7 families. Maybe it could be re-phrased to make it slightly clearer (if this is the correct meaning):

“The study population consisted of 3.5 million, which included 1.7 million families belonging...”

Also, what is meant by families? Household members sharing the same health insurance policy?

2. In supplementary table 1 – there is a ** after “Rest of the insured population” but no ** underneath the table.

Reviewer #2: COVID-19 hospitalization and mortality and hospitalization-related utilization and expenditure: Analysis of a South African private health insured population.

This paper looking at data from 188000 members of a health insurance in South Africa. The cross sectional study looked at the rates of hospitalisation and death of these insured people.

The value of the study would be to see if the risks, hospitalisation and mortality differed between the public and private sectors in SA.

In the background only relying on systematic reviews is limiting when it comes to studying Africa. Other papers exist and reports etc . for example:

Jassat, W., Cohen, C., Tempia, S., Masha, M., Goldstein, S., Kufa, T., Murangandi, P., Savulescu, D., Walaza, S., Bam, J. L., Davies, M. A., Prozesky, H. W., Naude, J., Mnguni, A. T., Lawrence, C. A., Mathema, H. T., Zamparini, J., Black, J., Mehta, R., Parker, A., … DATCOV author group (2021). Risk factors for COVID-19-related in-hospital mortality in a high HIV and tuberculosis prevalence setting in South Africa: a cohort study. The lancet. HIV, 8(9), e554–e567. https://doi.org/10.1016/S2352-3018(21)00151-X

Mendelsohn AS, De Sá A, Morden E, Botha B, Boulle A, Paleker M, Davies MA. COVID-19 wave 4 in Western Cape Province, South Africa: Fewer hospitalisations, but new challenges for a depleted workforce. S Afr Med J. 2022 Feb 1;112(2):13496. PMID: 35139985.

Van der Westhuizen JN, Hussey N, Zietsman M, et al. Low mortality of people living with diabetes mellitus diagnosed with COVID-19 and managed at a field hospital in Western Cape Province, South Africa. South African medical journal = Suid-Afrikaanse tydskrif vir geneeskunde 2021; 111(10): 961-7.

Phaswana-Mafuya N, Shisana O, Jassat W, et al. Understanding the differential impacts of COVID-19 among hospitalised patients in South Africa for equitable response. South African medical journal = Suid-Afrikaanse tydskrif vir geneeskunde 2021; 111(11): 1084-91.

The other issue missing in the background is that there was (during waves) hospital overcrowding (with private hospitals and /or public hospitals turning people away. – which would impact on mortality)

Costs are also relative and perhaps the Health Market Inquiry should be mentioned which describes over servicing in the private sector.

A spelling mistake on page 12 – Comorbidities

Methods: I am not able to comment on the statistical analysis and suggest that a statistician is consulted.

Results

It would be useful to compare the study population’s demographics to the general population.

Discussion

The hospitalisation and mortality rates vary according to the phase of the epidemic and the improvement in treatment.

It isn’t clear if diabetes with hypertension and heart disease has higher morbidity and mortality or whether it is diabetes with either one?

In paragraph 1 on page 16 – the HIV rates differ, is the number of H positive people too small? 4.7% seems still quite a number to me.– is there an assumption that those with HIV are more compliant? Is there evidence for this.

Page 16 paragraph 2 the other differences between public and private sector relate to testing and diagnosis.

Provincial differences needs more discussion – there are similar hospital groups in different provinces – are their systems different. Were there differences in private hospital access to oxygen, ICU specialists/ expertise, and overcrowding in different provinces?

The average length of stay appears to be longer in the national surveillance system, any ideas on this? It is unclear what exactly is included in the costing.

You should compare the costs to those found in Edoka et al.

Edoka I, Fraser H, Jamieson L, Meyer-Rath G, Mdewa W. Inpatient Care Costs of COVID-19 in South Africa’s Public Healthcare System. Int J Health Policy Manag 2021.)

Which policies will be affected by these results – are they substantially different to those in the public sector and the literature?

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

10 Mar 2022

Response to reviewer comments: PONE-D-21-33875

Reviewer 1

I think this is an important study which adds a lot of value, particularly as it covers a region of the world for which there has been limited information. The introduction is well written and interesting. It clearly identifies the research question and highlights the importance of this study. It is a rich dataset, which covers a long period of time, not just the beginning of the pandemic, allowing for changes in hospitalisation, mortality, hospital length and cost to be observed. However, I do have several concerns about the methodology which I have outlined below. Unfortunately, Tables 1-4 also seem to be missing from the manuscript, without which it makes it very difficult to assess the results. I have done my best without them, but would appreciate if a copy of the tables could be sent to me so that I can review the results properly. I also think there lacks interpretation and contextualisation of the results in the discussion (highlighted below).

Response: We apologise that you did not have access to the tables. Due to their size they were uploaded as a separate file which had to be accessed via a link in the article file. We have however now included them in the main article file.

I think if the authors could address these methodological concerns, and expand the interpretation in the discussion, it is a very worthwhile paper.

1. In your population, do you know all people who have received a positive test? Or is it possible that individuals could have received a positive test from a different source, and this is not recorded in the data? This is an important point and should be made clear in the paper. If it is possible that positive tests could have been missed, then this warrants further discussion (i.e. if those individuals did not need hospitalization, this would over-estimate the estimates of hospitalisation and mortality).

Response: As this data set is from a period prior to the mass distribution of home-based COVID-19 self-test kits, COVID testing was by doctor referral available under private insurance at no charge where the patient tested positive. For those paying out of pocket the test was relatively expensive (US$55 per test). It is possible that some patients with private voluntary insurance elected to seek testing in the public system or pay out of pocket for the test in which case they would be missed from the data set. However, we consider that this number would be very small as there was strong incentive for those insured to utilize their insurance benefits, and the requirement for a doctor referral also strengthened capture of data. We further note that (1) our study population was limited to those who tested COVID-19 positive and (2) all the statistics that we present are based on those who tested positive (not the entire insured population). Brief text has been added to the manuscript to reflect this point on page 11.

2. I think there are comorbidities that are known risk factors for COVID-19 which have not been included in the study.

You state in the methods:

“Conditions were considered as comorbidity risk factors for COVID-19 based on a review of published literature: Cancer, Chronic Renal Disease, Congestive Cardiac Failure, Chronic Obstructive Pulmonary Disease, Diabetes Mellitus, HIV, Hypercholesterolaemia, Hypertension, Hypothyroidism, Ischaemic Heart Disease, Pregnancy, Tuberculosis”

There are conditions that are considered to be risk factors for COVID-19 hospitalisation and death which are not included in this list, i.e. Asthma, chronic liver disease, neurological diseases (stroke/dementia/etc), organ transplant, immunosuppressive condition (rheumatoid arthritis/lupus/etc), in addition to others. For example, see this paper: https://www.nature.com/articles/s41586-020-2521-4#Sec2

Response: Thank you for the Williamson et al 2020 reference describing risk factors for COVID-related death that examined 17.3 million patient records in the UK NHS system. We undertook a thorough literature review and consulted with the South African-based medical experts overseeing utilisation management at the health insurance to decide on the final list of co-morbidity risk factors to include as well as consideration of the health profile of private sector patients in South Africa. We therefore do not consider that any major risk factors relevant to our population have been excluded. We have added text to clarify this process on page 6.

3. There are also risk factors such as smoking and obesity that have not been included presumably due to data limitations. It is mentioned in the discussion that obesity could not be included, but I think further discussion on impact of not including them and how this may have affected the results is warranted. (i.e. discussion of the smoking rates/obesity rates)

Response: Thank you for the comment and we agree that this is a limitation of the available data. As noted we do record this in the limitations section, and have added additional text in relation to smoking status (page 11)

4. Table 1 - 4 seem to be missing from the manuscript. I have done my best to assess the results without them, but it would be really helpful to see these tables.

Response: We apologise that you did not have access to the tables. Due to their size they were uploaded as a separate file which had to be accessed via a link in the article file. We have however now included them in the main article file.

5. It is also not completely clear why vaccination could not be included within this analysis. It would be useful to have more of a discussion on the effect of vaccination on hospitalisations and length of stay, and more information on vaccine roll out in South Africa (i.e. % of people vaccinated with 1 or 2 doses during the different waves), and how this would have affected the results observed.

Response: We have expanded the paragraph in the discussion on vaccination (page 11), and note that the data set timeframe did not correspond with the South African vaccine roll-out. While it would be useful to identify the impact of the vaccine on risk factors, we consider that reporting risks from a “pre-vaccination” timeframe as in our analysis will allow useful comparisons for future analysis.

6. I don’t think this is included in the Tables, but I would be interested to see a summary of the different comorbidities (i.e. XX reported Diabetes, XX Cancer; rather than just 0, 1 comorbidities). Perhaps this could be a summary table of the study characteristics. I appreciate this might be a lot of different risk factors and you may have small numbers for some, so perhaps the main comorbidities rather than 0,1,2,>3. Apologies if this is already included within the table 1.

Response: Apologies that you were not able to see the tables. Table 4 includes the major co-morbidities and common combinations of co-morbidities and their rates of hospitalisation and death.

7. Adjusted odds ratios are reported, but it is not clear from the methods exactly what you have adjusted by. This is also true for the reported adjusted IRR. I think this could be clarified in the methods (i.e. in the results you say that you did not adjust by medical insurance level for mortality risk as this was highly correlated with age and comorbidities, was this also the case for the hospitalization risk and length of stay/utilization?)

Response: For the multivariable model analyses, we included several well-known measured predictors (e.g. age, gender, province, hospital network, wave) as well health region random effect. Health region was chosen for the random effects to account for potential differences in the population served and the quality of care within different geographic health regions. In the analyses, we performed both univariate and multivariable analyses (including all the predictors and random effects) to identify independent predictors of the modelled outcomes. The multivariate analyses produced adjusted effects as opposed to unadjusted effects from using univariate analysis. This has been added to the analysis methods (page 7) and we have also amended the description of the methods in the abstract and modified reporting of results to help clarify.

8. I think one of the key strengths of this study is the longer time period over which data was collected, yet this is not really touched on in the discussion. It is not clear to me why mortality and hospitalization would be lower in wave 1? I would have thought that this was before vaccination, and so mortality might have been higher? I think some further discussion would be useful to help readers contextualise these results.

Response: Vaccination was not available in South Africa during the period included in this analysis. Hospitalisation and mortality risk were both higher in wave 2 compared to wave 1. This likely reflects the higher case load, health system pressures and the Beta variant in South Africa during wave 2 compared with wave 1. We have included these points in the discussion, page 10.

Some minor points:

1. This sentence about the study size in the methods is a little confusing:

“The study population consisted of families (1.7 million) and individuals (3.5 million) belonging to 19 health insurances administered by DH, representing around a third of South Africa’s privately insured

population.”

It is confusing as to whether it is individuals + families, or total individuals, and of these, there are 2.7 families. Maybe it could be re-phrased to make it slightly clearer (if this is the correct meaning):

“The study population consisted of 3.5 million, which included 1.7 million families belonging...”

Also, what is meant by families? Household members sharing the same health insurance policy?

Response: We have re written this sentence to make the study population clearer (page 5).

2. In supplementary table 1 – there is a ** after “Rest of the insured population” but no ** underneath the table.

Response: We have revised the column headings to make the table clearer.

Reviewer #2:

This paper looking at data from 188000 members of a health insurance in South Africa. The cross sectional study looked at the rates of hospitalisation and death of these insured people.

The value of the study would be to see if the risks, hospitalisation and mortality differed between the public and private sectors in SA.

In the background only relying on systematic reviews is limiting when it comes to studying Africa. Other papers exist and reports etc . for example:

Jassat, W., Cohen, C., Tempia, S., Masha, M., Goldstein, S., Kufa, T., Murangandi, P., Savulescu, D., Walaza, S., Bam, J. L., Davies, M. A., Prozesky, H. W., Naude, J., Mnguni, A. T., Lawrence, C. A., Mathema, H. T., Zamparini, J., Black, J., Mehta, R., Parker, A., … DATCOV author group (2021). Risk factors for COVID-19-related in-hospital mortality in a high HIV and tuberculosis prevalence setting in South Africa: a cohort study. The lancet. HIV, 8(9), e554–e567. https://doi.org/10.1016/S2352-3018(21)00151-X

Mendelsohn AS, De Sá A, Morden E, Botha B, Boulle A, Paleker M, Davies MA. COVID-19 wave 4 in Western Cape Province, South Africa: Fewer hospitalisations, but new challenges for a depleted workforce. S Afr Med J. 2022 Feb 1;112(2):13496. PMID: 35139985.

Van der Westhuizen JN, Hussey N, Zietsman M, et al. Low mortality of people living with diabetes mellitus diagnosed with COVID-19 and managed at a field hospital in Western Cape Province, South Africa. South African medical journal = Suid-Afrikaanse tydskrif vir geneeskunde 2021; 111(10): 961-7.

Phaswana-Mafuya N, Shisana O, Jassat W, et al. Understanding the differential impacts of COVID-19 among hospitalised patients in South Africa for equitable response. South African medical journal = Suid-Afrikaanse tydskrif vir geneeskunde 2021; 111(11): 1084-91.

Response: Thank you for alerting us to these articles. Three of these articles were published after we submitted our paper to Plos One (on 27 October 2021) and we had referenced another article from Jassat presenting data from the same cohort study (reference 14). We have now included reference to the suggested papers where we feel they have relevance.

The other issue missing in the background is that there was (during waves) hospital overcrowding (with private hospitals and /or public hospitals turning people away. – which would impact on mortality)

Response: Thank you for your comment. We agree that overcrowding has been a particularly relevant issue for the management of COVID-19, however this is a complex dynamic with uncertainty regarding the actual impact on death and COVID-19 outcomes and the distinction between delayed or deferred admission and/or admission to alternative facilities. Critically for this analysis, it is uncertain whether overcrowding had a differential impact on admission and death contingent upon particular risk factors. We consider that inclusion of the different pandemic waves in our models would incorporate the impact of overcrowding. We consider that this point is more suited for the discussion rather than the background and have add text in the discussion to reflect this point (page 11).

Costs are also relative and perhaps the Health Market Inquiry should be mentioned which describes over servicing in the private sector.

Response: We have made reference to the health market enquiry in the discussion in relation to private sector costs. We feel this is best placed in the discussion section (page 12).

A spelling mistake on page 12 – Comorbidities

Response: We have corrected this word throughout with no hyphen.

Methods: I am not able to comment on the statistical analysis and suggest that a statistician is consulted.

Response: Professor Manda is a statistician and has undertaken all of the statistical analysis.

Results

It would be useful to compare the study population’s demographics to the general population.

Response: We have expanded on the description in the background of the differences between the general population reliant on the public health sector and the private insured population in South Africa (page 4)

Discussion

The hospitalisation and mortality rates vary according to the phase of the epidemic and the improvement in treatment.

Response: we have included text in the discussion regarding differences in outcomes between pandemic waves (page 10).

It isn’t clear if diabetes with hypertension and heart disease has higher morbidity and mortality or whether it is diabetes with either one?

Response: thank you we have re written that sentence. It is the combination of diabetes, hypertension, hypercholesterolemia and ischaemic heart disease that carried the greatest risk of hospitalisation and mortality (page 10).

In paragraph 1 on page 16 – the HIV rates differ, is the number of H positive people too small? 4.7% seems still quite a number to me.– is there an assumption that those with HIV are more compliant? Is there evidence for this.

Response: The difference in HIV prevalence between patients attending the private and public sector in South Africa is well described and is related to underlying social and economic determinants of vulnerability to HIV. We have cited another study from South Africa that shows a similar difference between HIV prevalence amongst public and private sector patients admitted with COVID-19 (page 10)

Page 16 paragraph 2 the other differences between public and private sector relate to testing and diagnosis.

Response: We do not expect rates of testing in the public and private sector to have an influence on mortality amongst hospitalised patients. Certainly rates of testing would influence COVID-19 case numbers but should not influence mortality.

Provincial differences needs more discussion – there are similar hospital groups in different provinces – are their systems different. Were there differences in private hospital access to oxygen, ICU specialists/ expertise, and overcrowding in different provinces?

Response: we have added further discussion on possible reasons for the provincial differences in outcomes (page 10).

The average length of stay appears to be longer in the national surveillance system, any ideas on this? It is unclear what exactly is included in the costing.

Response: We have not found any published data on length of hospital stay in the national surveillance system that corresponds to the period that we considered. We have expanded on the description of costing data on page 6.

You should compare the costs to those found in Edoka et al.

Edoka I, Fraser H, Jamieson L, Meyer-Rath G, Mdewa W. Inpatient Care Costs of COVID-19 in South Africa’s Public Healthcare System. Int J Health Policy Manag 2021.)

Response: Thank you for the Edoka et al paper which we are aware of and which estimated hospitalization costs in the public sector using a bottom up/micro approach. The Edoka paper is a methodologically robust analysis and an important contribution to the literature but there is very little comparability between private sector provider fees charged to insurers and a bottom up estimation of “cost of provision” in the public sector. The reasons for this lack of comparability are important and are a unique area of research in costing methodology but are not the subject of this paper. Of particular importance within this paper is the variation of fees charged between providers for seemingly similar cases in the same context.

Which policies will be affected by these results – are they substantially different to those in the public sector and the literature?

Response: Thank you for the query. We have noted the various policy implications for the results including stratification by risk, reducing unwarranted variation and larger regulatory reform in terms of cost containment and improving efficiency. Some policy implications are implicitly targeted towards private sector (such as need for regulatory reform) but more general concepts relating to differentiated care based on risk are applicable to both public and private sectors. We are open to making revisions to text in specific instances to make this clearer and have reviewed the text but cannot identify any areas where this can be made more explicit.

________________________________________

Submitted filename: Response to reviewer comments Plos one.docx

Click here for additional data file.

21 Apr 2022

COVID-19 hospitalization and mortality and hospitalization-related utilization and expenditure: Analysis of a South African private health insured population.

PONE-D-21-33875R1

Dear Dr. Doherty,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Raymond Nienchen Kuo, Ph.D

Academic Editor

PLOS ONE

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: (No Response)

Reviewer #2: I acknowledge and am happy with the changes the authors have made. The paper is ready for publication.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

27 Apr 2022

PONE-D-21-33875R1

COVID-19 hospitalization and mortality and hospitalization-related utilization and expenditure:  Analysis of a South African private health insured population.

Dear Dr. Doherty:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Professor Raymond Nienchen Kuo

Academic Editor

PLOS ONE