Healthcare Financing Vulnerability and Service Utilization in Kenya During the COVID-19 Pandemic, with a Focus on Policies to Protect Human Capital

Abstract

The analysis of household health financing vulnerability and its impact on health service utilization during the COVID-19 pandemic remains inadequately explored in Kenya. This study was designed to examine the impact of health financing vulnerability on health services utilization during the COVID-19 period. A health financing vulnerability index (HFVI) was constructed to assess the financial risk that individuals faced in accessing essential health services. A pooled panel probit model was estimated to measure the effect of HFVI on service uptake. The study found a significant negative association between HFVI and health service utilization, indicating that a high level of health financing vulnerability is linked to poor health in periods of emergencies. To address this issue, the study recommends implementation of multiple policy measures during crisis periods, including enhancing social health insurance, providing financial support to vulnerable households, and increasing public expenditure on primary healthcare systems across counties, especially on drugs, referral logistics, personnel, medical equipment, and diagnostic technologies.

1. Introduction

1.1. Background

Healthcare in Kenya is provided through public, private-for-profit, and private not-for-profit facilities. Healthcare services are arranged in tiers running from level 1 (dispensary, the lowest level of care) to level 6 (referral hospitals, the highest level of care). Public health facilities are found in the lower levels of care, while private for-profit facilities are concentrated in the higher levels of care (Ministry of Health et al., 2005).

In the years since independence in 1963, the health financing landscape has been oscillating from a free to a user-charge system (Chuma & Okungu, 2011; Wamai, 2009; Chuma et al., 2009; Mwabu et al., 1995). Currently, healthcare in Kenya is financed from three main sources: out-of-pocket charges, tax revenues, and donor funds. In 2005–06, out-of-pocket payments (OOP) were 29.1% of total health expenditure (Ministry of Medical Services & Ministry of Public Health and Sanitation, 2009b). Large out-of-pocket payments can be a barrier to sustainable healthcare because over time, they make individuals vulnerable to poverty, or frustrate the exit from poverty (Ogwang & Mwabu, 2025). A survey performed in 2007 showed that 38% of people who were ill cited a lack of money as a barrier to seeking healthcare (Ministry of Medical Services & Ministry of Public Health and Sanitation, 2009a). By 2009–10, out-of-pocket payments still made up nearly a quarter of total health expenditure. The adverse impact of user fees on healthcare utilization in Kenya was brought to light when fees for maternity services at dispensaries and health centers were abolished (Leftie, 2013), resulting in a massive influx of patients to these facilities (Stacey et al., 2021).

Briefly, Kenya’s health system is financed through four main sources: government revenue, out-of-pocket charges, foreign assistance, and private sector contributions, estimated, respectively, at 45.98%, 24.3%, 18.51%, and 35.51% of the total health expenditure (Ministry of Health, 2021).

Figure 1 provides the trend of the Kenyan government budgetary allocations to the health sector for the period 2012–21.

1.2. Research Problem

The seriousness of the healthcare financing gap in Kenya was revealed by the advent of the COVID-19 pandemic in 2019–22, which found the country unprepared for the sudden health needs of the population. The pandemic precipitated distress in both urban and rural health facilities, as these units did not have the capacity to care for patients, especially those with COVID-19 symptoms (Puro & Kelly, 2021). The situation forced the Government to reallocate, in favor of COVID-19 emergencies, scarce health resources already committed to tackling existing health conditions such as HIV, TB, Malaria, and childhood diseases (Barasa et al., 2021).

During the COVID-19 pandemic, utilization of healthcare services in health facilities registered a significant drop in practically all 47 Kenyan counties (Wambua et al., 2022). For example, there were substantial declines in clinic attendance during the month of April 2020. In particular, drops in visits were observed for all under- and over-five children, for final antenatal care, for hypertension and diabetes, and for persons needing HIV testing.

The factors associated with the declines in visits included (i) the fear of contracting the various variants of the COVID-19 disease, (ii) inability to access healthcare services due to lack of income, and (iii) waiting time due to large crowds at health facilities (Wambua et al., 2022). Vulnerability to a lack of care for any illness during the COVID-19 period was compounded by the policies undertaken by the government to contain the pandemic. The containment measures included imposition of strict curfews—lockdown of schools, businesses, and marketplaces, banning of international flights, local travel, and of large social gatherings, among others (Onditi et al., 2020).

The lockdowns caused varied economic consequences, such as job loss, food and housing insecurity, further aggravating the health crisis (Onditi et al., 2020). Thus, it is not surprising that during the COVID-19 pandemic, utilization of healthcare services at both public and private facilities significantly went down. However, the extent to which the drop in healthcare utilization is linked to households’ economic vulnerability associated with the pandemic has not hitherto been investigated. An important facet of this study was to evaluate the effect of household healthcare financing vulnerability on the utilization of healthcare during the COVID-19 period. A related objective was to use the evidence generated to suggest policies to protect human capital, particularly health capital, in contexts of disease pandemics and emergencies.

2. Related Literature

Initial studies of the effect of COVID-19 on utilization of health services showed widespread reductions (Riley et al., 2020), as seen in previous epidemics in Sub-Saharan Africa (Barden-O’Fallon et al., 2015; Bolkan et al., 2018; Brolin Ribacke et al., 2016; Elston et al., 2016; Takahashi, 2015). What is also notable from previous studies is that some population groups are at greater risk of being severely affected during periods of disease outbreaks, particularly children, women, and persons in extreme poverty (Barden-O’Fallon et al., 2015).

In Kenya and Uganda, there is evidence that COVID-19 caused more than two-thirds of households to experience adverse income shocks, thus reducing their ability to afford food and health services (Kansiime et al., 2020). This evidence finds support in Kang et al. (2023) in Chad, who found that two-thirds of households, both rural and urban, reported income reductions, with urban areas being hit hardest in 2020, and the rural areas suffering similarly in 2021. Solymári et al. (2022) shows that COVID-19 was associated with large reductions in health service utilization, particularly in slum areas. This study is closest to that of Janssens et al. (2021), regarding the negative effect of COVID-19 on healthcare uptake among low-income households. Janssens et al. found that 80 percent of their sample had experienced income loss during the COVID-19 period. Edeh et al. (2025) show that COVID-19 reduced catastrophic health expenditure in Nigeria, since, arguably, legal restrictions limited access to medical services, thus reducing out-of-pocket expenditure. This finding was in contrast to Jung et al. (2021), who found that the loss of household income created an economic burden, thus increasing catastrophic healthcare spending.

Bruce et al. (2022) report financial hardships among American households compounded by worries that COVID-19 was aggravated by existing health conditions and co-morbidities. The study found that the households that were financially constrained to seek healthcare pre-COVID-19 were the most affected by illnesses contracted during the pandemic period, while those in the middle class were neutrally affected, with some escaping the fangs of the pandemic. The findings of this study provided the State Governments in the United States with the evidence they needed to target help to vulnerable households, thereby reducing the risk of losing human capital and livelihoods. Similarly, efforts should be made in Africa to provide such evidence and further encourage its use in contexts of disease epidemics.

Many studies in low- and middle-income countries report negative associations between COVID-19 and health service utilization. See, e.g., Wambua et al. (2022), Siedner et al. (2020), Barasa et al. (2021), Hategeka et al. (2021), Burt et al. (2021), Dorward et al. (2021), Adelekan et al. (2020), Wong et al. (2021), and Zar et al. (2020).

In Kenya, Macharia et al. (2020) created COVID-19 vulnerability indices to identify Kenyan households who were at risk of not accessing healthcare during the pandemic. The authors constructed three indices spanning 295 sub-counties in the country. The indices included the Social Vulnerability Index (SVI), Epidemiological Vulnerability Index (EVI), and a Social Epidemiological Vulnerability Index (SEVI). The results found that 49 sub-counties in the north-western and eastern parts of Kenya had approximately 6.9 million vulnerable people, with only 58 sub-counties with 9.7 million people (out of a population of about 52 million) who were unlikely to be affected by the pandemic. These indices demonstrated county-level heterogeneities in welfare vulnerabilities occasioned by COVID-19. However, Macharia et al. (2020) did not analyze how the pandemic affected households’ ability to pay for healthcare. In particular, the link between health service utilization and households’ financial vulnerability, resulting from the COVID-19 pandemic, was not investigated. Households’ capacity to pay for healthcare was at risk of being eroded by COVID-19 itself and by the measures to contain it.

Specific studies in Africa evaluating the impact of lockdown measures to fight COVID-19 in South Africa detected a substantial drop in primary healthcare utilization (Adelekan et al., 2020; Siedner et al., 2020). In Kinshasa, Democratic Republic of Congo, significant drops in usage of essential health services were experienced following the adoption of public health measures against COVID-19 (Hategeka et al., 2021). However, the role played by COVID-19-related healthcare-financing vulnerability was not examined.

Various factors have been put forward to explain the downward trends in access to specific healthcare services during the COVID-19 pandemic across countries. The Partnership for Evidence-Based Response to COVID-19 (2020) conducted a survey in Kenya to assess determinants of health services utilization during COVID-19. The authors reported that the main factors hindering service uptake included fear of risk of catching coronavirus at healthcare facilities, reduced incomes, transport costs, health-related costs, shortage of health workers, and difficulties in accessing health facilities.

Shimeles et al. (2021) reported that low utilization of healthcare services during COVID-19 was driven by unavailability and unaffordability of care, increased prices of medications, and interruptions in follow-up visits. Aiello et al. (2008) and Liang et al. (2020) argued that pre-existing challenges in access to health services, such as poor road networks, disruptions in supplies to health facilities, and limited or lack of capacity for additional domestic healthcare financing. Morbidity and mortality outcomes of reductions in attendance were also studied (Apicella et al., 2020). However, none of the studies conducted during the COVID-19 period linked reductions in clinic attendance to healthcare financing vulnerabilities prevailing at the time.

3. Data, Methodology, and Analytical Frameworks

3.1. Data Sources

The paper utilizes high-frequency phone survey data collected to examine the socio-economic impacts of COVID-19 in Kenya. These data were collected through a collaborative effort involving the World Bank, the Kenyan National Bureau of Statistics (KNBS), the United Nations High Commissioner for Refugees (UNHCR), and the University of California, Berkeley. The survey was conducted in five waves, namely: (i) Wave 1: 14 May–7 July 2020; (ii) Wave 2: 16 July–18 September 2020; (iii) Wave 3: 18 September–28 November 2020; (iv) Wave 4: 15 January–25 March 2021; and (v) Wave 5: 29 March–13 June 2021. The sample for this study covers urban and rural areas and is generally representative of the population of Kenya. The second sample comprises households selected using the Random Digit Dialing method. A list of random mobile phone numbers was created using a random number generator from the 2020 sampling frame produced by the Kenya Communications Authority. The main categories of the questionnaire were designed to collect data on household background, travel patterns and interactions, employment, food security, transfers, subjective welfare, health, COVID-19 knowledge, household structures, and social relations.

3.2. Theoretical Model

According to Thomas (2021) and related literature, the demand for healthcare can be influenced by health needs, household means, and the standards and resource requirements of medical practice. Following Grossman’s (1972) health production model, as adapted from Becker (1965), individuals allocate their resources and time to investments that increase health and other consumption goods. The benefits of improved health include enhanced future utility flows, as well as increased time available for market and non-market production. The former implies that health is a consumer good, while the latter views health as capital (a factor of production). Further, the model observes that health is a commodity that differs from other goods and services in that individuals cannot purchase health but rather, must produce it themselves using purchased inputs, complemented, as necessary, by appropriate personal behaviors, such as exercising and abstinence from substance abuse. Thus, demand for healthcare is derived from demand for health (Hren, 2012).

Grossman’s model has received theoretical criticisms, including its failure to address uncertainty and the assumption that an individual has perfect information concerning the decision to maximize utility from health. The model is also criticized due to its assertion that it is produced by an individual, while in reality, the majority of individuals live within families. Moreover, this model does not analyze children’s demand for health and healthcare, and further fails to take into consideration the fact that an individual’s demand for healthcare can be influenced by the health needs of other family members (Hren, 2012). Consequently, many models that extend Grossman’s original work are concerned with the above issues. In some of the models, parents are assumed to maximize a family’s utility, rather than an individual’s (Rosenzweig & Schultz, 1983).

Despite these criticisms, the Grossman model provides valuable insights into the complexities of health production and health services utilization. The model particularly illustrates well the interlinkages between three commonly studied aspects in healthcare research, namely, the (i) demand for health, (ii) health service utilization, and (iii) health insurance uptake (Hren, 2012). The model is applied to the research issue of interest in this paper, taking into consideration innovations in related health service utilization studies—see, especially (Rosenzweig & Schultz, 1983).

3.3. Empirical Model

We estimate a healthcare utilization model in which the main driver of utilization is healthcare financing vulnerability rather than the usual user charges or an individual’s income (Muriithi & Mwabu, 2014). This is performed by first constructing a Healthcare Financing Vulnerability Index (HFVI) and then using this index as the main regressor in a structural model of health service utilization. HFVI is a composite index of factors that are negatively correlated with health services utilization, such as consultation fees, lack of income, or an adverse shock. In this study, HFVI is operationalized as the financial inability to pay for healthcare. Thus, HFVI has elements of high user charges at health facilities, household income poverty, adverse shocks of any kind, plus negative effects of a health system designed with insufficient regard to the economic contexts of individuals and households. It is also worth highlighting that HFVI is correlated with unobservable factors for which data is not available. However, the estimation method we use—the control function approach—partially accounts for this problem via the inclusion of the reduced-form residual in Equation (3) below.

We assume that an individual is healthcare financing-vulnerable if the person is less likely to afford the available healthcare, and consequently does not visit a health facility, or, conditional on the visit, cannot raise the required user charge, and hence returns home without treatment. As already noted, HFVI is linked to the health system. Health financing, as a function of a national health system, is “concerned with the mobilization, accumulation, and allocation” of resources to cover the cost of people’s health maintenance—see especially OECD (2013). The health system’s health financing function should ensure that the cost of care available to people is not beyond their means.

The HFVI is constructed using the variables that increase the risk of individuals being unable to use healthcare services provided by the health system. It is worth noting that a household’s accumulation and mobilization of funds to pay for healthcare can be undermined by a health system’s financing function that does not take into account the economic context of households. A hypothetical case is a health system that does not take into account the fact that in some communities, the majority of households can afford neither user charges nor insurance premiums or contributions.

In the construction of the HFVI, two methods were used, namely, the Multiple Correspondence Analysis (MCA), as modified by Deng and Tian (2015), and the Uncentered Principal Component Analysis (UPCA), as articulated in Wittenberg and Leibbrandt (2017). Both methods were used to construct a composite index based on a set of dummy variables that proxy the ability to pay. Wittenberg and Leibbrandt constructed and compared three asset indices, namely, the Usual PCA, MCA, and Uncentered PCA, where the dummy variables are given a value of zero, notably for asset possession, and a value of one, otherwise. The usual PCA and MCA had some negative values for asset indices that were undesirable on the asset scale. Wittenberg and Leibbrandt (2017) used Uncentered PCA, which did not have negative values, an approach adopted in this study.

Table 1. The Components of Healthcare Financing Vulnerability Index (HFVI).

Variable	Description	Reason
Employment	A dummy variable taking the value 1 for not being employed, and a value of zero otherwise.	Persons who are not employed are more vulnerable to the inability to afford health care finance than those who are employed (Bruce et al., 2022; Kitole et al., 2023).
Employment type	A dummy variable taking the value 1 for part-time employment, zero for full employment.	Those in part-time employment are more vulnerable to lack of funds to pay for health care compared to those in full-time employment (Kitole et al., 2023)
Health insurance	A dummy variable taking the value 1 for not having insurance, zero otherwise;	Persons without health insurance are more vulnerable to inability to pay for health care, a situation that increases HFVI (Kitole et al., 2023)
Employer’s medical benefits	A dummy variable taking the value 1 for not having the employer’s medical benefits, 0 otherwise;	The households without medical benefits are more vulnerable to inability to use care (Kitole et al., 2023)
Cash payments	A dummy variable taking the value 1 for not paying for health services in cash, and 0 otherwise.	Persons paying in cash are less vulnerable to inability to finance health care (National Academies of Sciences, Engineering, and Medicine, 2018; Houeninvo et al., 2023)

Table 1 shows that HFVI is a composite index of indicators that proxy for a high inability to afford healthcare. In the process of aggregating data through Uncentered PCA, the selected variables were adjusted to exhibit positive correlations with the HFVI concept. This adjustment ensures that the variables are combined effectively using Uncentered PCA and exhibit the desired positive correlation with the inability to pay for healthcare.

shows the dummy variables used to construct the HFVI, as well as the codes for the dummies.

The HFVI has been constructed at both the national and county levels, highlighting the counties that exhibited greater susceptibility to healthcare financing challenges. These findings are visually presented on a map of Kenyan counties in the results section.

The HFVI (Hfv) is the main driver of healthcare utilization since it is constructed using the variables representing a household’s inability to pay for care (see Equation (1)). Other determinants of service utilization include cost of medical services, cost of medical equipment, cost of food items, and demographics, especially age, gender, marital status, education, occupation, household size, pre-existing health conditions, residence, and distances to sources of care.

We use a pooled panel probit approach to estimate the parameters of our healthcare utilization equation. The pooled panel model is less restrictive than the longitudinal probit model, but it works best when serial dependence is absent or can be handled by clustering standard errors, although it does not explicitly control for time-invariant unobserved heterogeneity. Assuming there is no autocorrelation in our dataset, using a pooled probit model may offer a better fit to the data than a longitudinal model that does not account for time-invariant unobservable covariates—see, Wooldridge (2002, pp. 482–485).

We estimate a pooled panel probit model of the form:

$$Uh{c}_{it}=\mu +\eta Hf{v}_{it}+\beta Z_{it}+u_{it}$$

(1)

where

$Uh{c}_{it}$ = Utilization of healthcare services proxied by whether any member of the household visited any health facility in the past 7 days.

$Hf{v}_{it}$ = Health financing vulnerability index constructed from Table 1.

$u_{it}$ = Random error term.

The coefficient η, on $Hf{v}_{it}$ indicates the impact of Health Finance Vulnerability Index on healthcare utilization. Since $Uh{c}_{it}$ and $Hf{v}_{it}$ are likely to be jointly determined, we address this endogeneity issue by instrumenting $Hf{v}_{it}$ by the non-self-mean of $Hf{v}_{it}$ for every county (see Strauss, 1986; Epo et al., 2025). The reduced form equation for the Health Financing Vulnerability Index is:

$$Hf{V}_{it}^{\ast }=\mu +\beta Z_{it}+\sigma I{V}_{it}+{\epsilon }_{it}$$

(2)

where

$Hf{V}_{it}^{\ast }$ = Health Financing Vulnerability Index constructed using a composite index from the inability to pay index.

$Z_{it}$ = vector of exogenous covariates.

$I{V}_i$ = the non-self-mean of the Health Financing Vulnerability Index for each of the 47 counties, such that all households in a given county have the same value for this instrumental variable.

${\epsilon }_{it}=r_i+u_{it}$ = a new composite error, where r is fixed, and u is an idiosyncratic term.

To address heterogeneity emanating from the non-linear interaction of HFVI with the unobservable variables in the error term, we construct an interaction variable between HFVI with the predicted residual of the HFVI as an additional regressor to obtain the following expression:

$$Uh{c}_{it}^{\ast }=\alpha +\lambda Hfv{i}_{it}+\beta Z_{it}+\gamma Hfviresi{d}_{it}+\pi {(Hfviresid\ast Hfvi)}_{it}+{\epsilon }_{it}$$

(3)

The residual ($Hfviresid$) serves as a control function variable that renders HFVI exogenous (Wooldridge, 2015; Mwabu, 2009). As noted, the interaction term $(Hfviresid\ast Hfvi)$ addresses the unobserved heterogeneity of the coefficient on HFVI, keeping it constant across units of analysis, ${\epsilon }_{it}$ is a composite error term comprising the random and the nonrandom, fixed part of the error term, and $\mathsf{\alpha }, \mathsf{\lambda }, \mathsf{\beta }, \mathsf{\gamma }, \mathrm{and} \pi$ are vectors of parameters to be estimated.

Table 2. Definition of Variables.

Variables	Measurement
Dependent Variable
Utilization of health care services	A dummy variable taking the value 1 if any member of the household visited any health facility in the past 7 days, 0 otherwise
Explanatory Variables
Health Financing Vulnerability Index (HFv)	This is a continuous variable constructed from dummy variables in Table 1 using Uncentered PCA
The Z vector in Equation (3) contains: The natural logarithm of the cost of medical services, including medical and non-medical costs	A continuous variable indicating the natural logarithm of the total amount in Kshs paid for the item(s)/service(s) used in service delivery
The natural logarithm of the cost of other related commodities, such as food	A continuous variable indicating the natural logarithm of the total amount in Kshs paid for food items
Marital Status of an individual	A dummy variable that takes the value 1 if one is married, 0 otherwise
Residence	A dummy variable taking the value 1 if an individual resides in an urban area, 0 otherwise
Gender	A dummy variable taking the value 1 if an individual is male, 0 otherwise
The natural logarithm of the age	The natural logarithm of the age of an individual in years
Primary Education Level	A dummy variable taking the value 1 if an individual has primary education, 0 otherwise
Secondary Education Level	A dummy variable taking the value 1 if an individual has secondary education, 0 otherwise
University Education Level	A dummy variable taking the value 1 if an individual has a university education, 0 otherwise
The natural logarithm of the household size	A continuous variable indicating the natural logarithm of the total number of household members
Pre-existing health conditions	A dummy variable taking the value 1 if there is a pre-existing health condition, such as diabetes, etc., 0 otherwise
Instrumental Variable (IV)	This is a non-self-county mean of HFVI constructed by getting the county mean for all HFVIs, except for the index of individual i, so that the mean is exogenous to individual i.
Predicted residual of HFVI, i.e., $HFVIresi{d}_{it}$	This is a continuous variable generated from a reduced-form regression of the potentially endogenous HFVI on all exogenous variables and the instrumental variable
Interaction of the predicted residual with HFVI, i.e., $(HFVIresid \ast HFVI{)}_{it}$	This is a continuous variable generated by interacting the predicted residual of HFVI with the potentially endogenous HFVI

provides the definition and measurement of the variables used in our model estimation.

3.4. Study Limitations

In interpreting the evidence presented in Section 4 or in using it to inform policy, the following considerations should be borne in mind. The dependent variable in the empirical model, i.e., health service utilization in the past 7 days, might be incorrectly measured because some current health conditions can necessitate health service utilization over a long time period. Moreover, in measuring health service utilization over the last 7 days, account is not taken of the frequency of service utilization. Briefly, health service utilization over the period allowed in the survey is likely to be understated. Another methodological shortcoming is the possibility of sampling bias due to the fact that the analytic sample was based on respondents who possessed a mobile phone. This was because of the legal restrictions during the COVID-19 pandemic that prevented the usual survey design that involves some form of physical interaction. However, even in a physical setting, some sampling bias might not be avoidable. The estimated coefficient on HFVI should be interpreted with caution because the endogeneity problem might not have been fully addressed due to data limitations. The estimates have a causal interpretation to the extent that the assumptions regarding the pooled probit model are valid. Furthermore, the unobservable variables that drive health service utilization can cause a correlation between the structural and reduced-form disturbance terms in Equations (1) and (2), thus biasing the coefficient on HFVI. Finally, as again pointed out by a reviewer of this journal, we have no way to test the hypothesis that our instrumental variable is properly excluded from the structural equation because it is the only one available, and is thus assumed to influence health service utilization strictly through HFVI.

4. Results

The research findings are presented in three sections. The first part deals with the descriptive statistics of the variables of interest, which are used in the regression analyses. It also provides the distribution of the health financing vulnerability index (HFVI) at the county level, which shows the spatial distribution of the index. Part two of the findings provides estimation results, including the effect of HFVI on the utilization of healthcare during the COVID-19 period. The third part contains policy recommendations drawn from the findings.

4.1. Descriptive Statistics

Table 3. Summary Statistics for Healthcare Services Utilization (n = 59,838).

Variables	Mean	Standard Deviation	Min	Max
Healthcare Services Utilization	0.1616	0.3681	0	1
Health Financing Vulnerability Index (HFVI)	1.72	7.0955	0	5.1925
Natural log of price of medical services in Kshs	6.0376	1.2294	0	11.5129
Natural log of the prices of food items in KSh	6.9650	0.7855	2.3026	11.2252
Married (=1)	0.1160	0.3202	0	1
Area of residence (rural area = 1)	0.5218	0.4995	0	1
Gender (man = 1)	0.4795	0.4996	0	1
Natural log of age of the respondent	3.5217	0.3975	2.8904	4.7005
Primary education level (=1)	0.3008456	0.4586	0	1
Secondary education level (=1)	0.4357432	0.4958	0	1
University education level (=1)	0.0337578	0.1806	0	1
Pre-existing health condition (=1)	0.0010863	0.0329	0	1
Natural logarithm of the Household size	1.456351	0.5740	0	3.2958
County level Mean of Health Financing Vulnerability Index (Instrumental Variable)	1.4	0.0643	0	3.9
Predicted residual of HFVI	−9.14 × 10−10	5.368562	−4.65507	51.56799
Interaction of the predicted residual with HFVI	28.81862	211.0986	−5.098032	2651.902

Source: Author’s own computation.

presents the descriptive statistics for the variables used to estimate the utilization of healthcare services during the period of COVID-19 in Kenya. The key variable of interest is the Health Finance Vulnerability Index at the county level. The other variables are control variables that are used in the healthcare utilization model. The national-level mean score for the health financing index was high, registering at 1.72, relative to a maximum of 5.2 with a median of 1.8. This indicates that, in general, Kenyan households faced a notable degree of vulnerability in terms of health financing. Indeed, it should be noted that a score near zero suggests low vulnerability, in contrast to scores greater than one. The higher the score, the greater the vulnerability.

From the descriptive statistics, 16% of the respondents sought medical care during the COVID-19 period. This contrasts with the expectation that COVID-19 would increase utilization of healthcare services, fueled by the perception that people would fall ill during the pandemic and seek treatment. It was widely reported that health facilities would be under significant strain during COVID-19. Instead, the literature indicates a decline in healthcare utilization, especially at the lower levels of the health system. Additionally, certain variables of interest, such as the prices of personal healthcare and related goods like food, exhibited a wide standard deviation, signifying substantial price variations across counties.

4.2. Spatial Distribution of Health Financing Vulnerability Index

The HFVI assesses a household’s financial capacity to cope with a severe health shock by visiting health facilities for treatment, with lower financial capacity, indicating little ability to cope. The higher the value of the index, the more severe the household’s inability to finance a visit to a health facility. The mean of HFVI at the country level was 1.72.

The means are shown in Figure 2. In the corresponding spatial mapping, the deep red indicates counties with the highest HFVI. These correspond to the majority of the counties in the northern part of Kenya, with a HFVI as high as 2 (Figure 2). Looking through the bar graphs in Figure 2, Marsabit County leads with a HFVI of 2.6, followed by Samburu with 2.3, while Turkana and Garissa rank third with 2.2. Other counties with HFVI of 2 and above included Isiolo, Tharaka-Nithi, West Pokot, and Bomet. Narok and Taita-Taveta have the lowest HFVI of 1.3. What is notable is that all the counties have HFVI greater than one, which suggests that all households are highly financially vulnerable in terms of being able to use healthcare when needed.

The distribution of HFVI tends to follow some trends that mimic the poverty levels in Kenya. The counties in the northern parts of Kenya show almost similar poverty levels, with a poverty incidence of more than 50% Kenya National Bureau of Statistics (2020). Of course, Mandera, which is located in the northern part of Kenya, has exhibited a lower HFVI of about 1.39, among the lowest indices in the country. The Kenya sub-national spatial distribution of the HFVI is shown in Figure 3.

Again, given that Mandera’s poverty incidence stands at around 77.6% as per the Kenya National Bureau of Statistics (2020), there is a likelihood of a sample selection bias given that the survey was conducted by use of mobile phones. Those with phones were likely to be well endowed with financial resources, which could have led to a misleading result that shows Mandera having a low HFVI.

A notable observation is that counties that had health insurance coverage, such as Makueni and Kitui, reported a low HFVI, meaning that health insurance was, to some extent, well identified as a good measure of constructing the HFVI. Again, although Nairobi County had an absolute poverty rate of around 17%, it had an HFVI of about 1.78, quite close to the county average of 1.72. Most likely, the above country-average index for Nairobi county (capital city of Kenya) could be due to the heterogeneous nature of poverty and inequality occasioned by the presence of large informal settlements (Shifa & Leibbrandt, 2017).

4.3. Drivers of the Utilization of Healthcare Services

The regression results for the healthcare utilization model are shown in

Table 4. Utilization of healthcare services during COVID-19: Control Function Estimates (n = 59,838).

Explanatory Variables	First Stage Regression: Reduced-Form Estimates Dependent Variable: Health Financing Vulnerability Index (HFVI)	Second Stage Regression: Control Function Estimates Dependent Variable: Healthcare Utilization (=1)
Health Financing Vulnerability Index	---	−0.8088 (0.0474)
Log Total Cost of Medicines	−0.1849 *** (0.0445)	0.2673 *** (0.0201)
Log Total Cost of Food items	0.3071 *** (0.0629)	0.4924 *** (0.0201)
Married (=1)	0.8093 *** (0.1829)	0.9239 *** (0.0581)
Area of Residence Rural area (=1)	0.0659 (0.1074)	0.0209 (0.02630)
Gender Male (=1)	0.1399234 (0.1074)	0.0669122 *** (0.0269)
Log Age of the Respondent	0.6537 *** (0.1384)	0.5632 *** (0.0449)
Primary (=1)	0.22618 (0.1485)	0.0078 (0.0375)
Secondary level (=1)	0.1810 (0.1465)	0.0071 (0.0362)
University level (=1)	0.6864 ** (0.3241)	0.4343 *** (0.0837)
Pre-existing Conditions (=1)	2.175815 *** (1.0368)	2.7923 *** (0.2920)
Log of Household Size	−0.7695 *** (0.1053)	−0.3541 *** (0.0450)
Non-self County Mean for HFVI (×10)	4.081 *** (0.0368)	-----
Predicted HFVI residual	-----	0.8076 *** (0.0470)
Interaction of predicted residual with HFVI	-----	−0.0003 (0.0002)
Constant	−1.3141 * (0.7147)	−4.2716 *** (0.1971)
Diagnostic Statistics	F(12, 59838) = 14.27 (p-value: 0.000) R-sq. = 0.0167 Adj R-sq = 0.0155 Root MSE = 5.3736	LR Chisq (14) = 1301.49 (p = 0.000) Prob > chi2 (p = 0.0000) Pseudo R2 = 0.0953 Log likelihood = −6176.41

Note: The figures in parentheses are the standard errors; while, * p ≤ 0.1; ** p ≤ 0.05; *** p ≤ 0.01. Source: Authors’ own computation.

.

The first stage model is an OLS regression, where HFVI is conditioned on its instrument—the non-self-cluster mean for HFVI (Strauss, 1986; Mwabu, 2009; Epo et al., 2025), with cluster being a county. The non-self-cluster mean affects the HFVI of every household. However, since, by design, the reverse does not apply, the instrument is exogenous.

The second stage regression reports the estimated effect of HFVI on the Probit index for healthcare utilization, controlling for the Z covariates in Equations (1) and (2). The Durbin–Wu–Hausman (DWH) test shows HFVI is endogenous, so the control function procedure is justified. The F-statistic for first-stage regression is 24.43, which is much higher than the threshold value of 10, showing that the instrument is exogenous (Wooldridge, 2015).

Table 5. Marginal Effects: Changes in Pr (Heath Service Utilization) associated with unit changes in the RHS covariates. Predicted Pr (Heath Service Utilization = 0.396.

Variables	Probit Marginal Effects
Healthcare Financing Vulnerability Index (HFVI)	−0.3117 *** (0.0182)
Log of medical input costs	0.1100 *** (0.0068)
Log the cost of food items	0.1898 *** (0.0078)
Married (=1)	0.3539 *** (0.0199)
Area of residence, Rural (=1)	0.0080 (0.0101)
Gender, Male (=1)	0.0258 *** (0.0104)
Log Age of the respondent	0.2171 *** (0.0173)
Primary level (=1)	0.0030 (0.0142)
Secondary level (=1)	0.0027 (0.0135)
University level (=1)	0.1716 *** (0.0329)
Pre-existing conditions (=1)	0.6006 *** (8.46)
Log Household size	−0.1365 *** (0.0173)
HFVI_predicted residual	0.3113 *** (0.0182)
Interaction of HFVI with predicted residual	−0.0001 (0.0001)

Notes: *** p ≤ 0.01; standard errors are in parenthesis. Source. Authors’ Own Computations.

shows the marginal effects corresponding to the probit coefficient estimates shown in Table 4. The marginal effects show the associations between the RHS covariates with the probabilities of visiting health facilities. Notice that the Right-Hand-Side (RHS) covariates in Table 4 include the variable of policy interest (the actual or observed HFVI) plus the associated control variables, i.e., HFVI_predicted residual, and the interaction of the predicted residual with HFVI.

Table 5 shows that the Healthcare Financing Vulnerability Index has a strong negative association with the probability of health services utilization. A unit increase in the index reduces the probability of seeking healthcare by 0.311 or by 31.1%.

This result is not attributable to what has been found in the literature by Aiello et al. (2008) and Liang et al. (2020). In

Table 6. Marginal Effects: Pr (Heath Service Utilization = 1); Predicted value = 0.396).

Variables	Probit Marginal Effects
Health Financing Vulnerability Index (HFVI)	−0.3117 *** (0.0182)
Log of medical input costs	0.1100 *** (0.0068)
Log the cost of food items	0.1898 *** (0.0078)
Married (=1)	0.3539 *** (0.0199)
Area of residence, Rural (=1)	0.0080 (0.0101)
Gender, Male (=1)	0.0258 *** (0.0104)
Log Age of the respondent	0.2171 *** (0.0173)
Primary level (=1)	0.0030 (0.0142)
Secondary level (=1)	0.0027 (0.0135)
University level (=1)	0.1716 *** (0.0329)
Pre-existing conditions (=1)	0.6006 *** (8.46)
Log Household size	−0.1365 *** (0.0173)
HFVI_predicted residual	0.3113 *** (0.0182)
Interaction of HFVI with predicted residual	−0.0001 (0.0001)

Notes: *** p ≤ 0.01; standard errors are in parenthesis. Source. Authors’ Own Computations Using the World Bank (2020) COVID-19 Data.

, pre-existing challenges in access to health services, such as poor road networks, disruptions in supplies to health facilities, and limited or no capacity for domestic production of medical supplies, are held constant. However, if they were to be included in our structural model, they might also have negative associations of their own with the probability of health service utilization. The same applies to inclusion in our model of variables that capture the preventive effects of hand hygiene and face masks.

Other control variables that are associated with notable results are the costs of medical services and related goods, such as the expense of food. Both are positively correlated with the probability of visiting a health practitioner during the COVID-19 period, which is consistent with them being proxies for income.

Other variables that are positively correlated with the probability of the household seeking healthcare during the COVID-19 pandemic include being married, which is consistent with the literature (Damiano et al., 2021).

5. Conclusions and Policy Implications

The study has measured the impact of HFVI on the utilization of healthcare facilities in Kenya during the COVID-19 pandemic. The econometric evidence shows a consistent and statistically significant negative association between healthcare financing vulnerability with health service utilization. The measures taken by the government to contain the pandemic seem to have worsened households’ inability to afford healthcare. Lockdown measures plus COVID-19-related sicknesses, together or separately, led to job losses and likely increased people’s inability to afford medical insurance or travel costs to sources of care. Appropriate fiscal measures and social transfers in times of disease pandemics can substantially reduce healthcare financing vulnerabilities in the population and protect both health and livelihoods. Moreover, since health is intrinsically connected to schooling and learning, such policies can further protect the formation of education capital in periods of disease pandemics. We have shown that health financing vulnerability index varies substantially by region, with counties in the north-eastern part of Kenya having the highest vulnerability indices, which suggests that geographic targeting of both fiscal benefits (e.g., exemptions from some tax categories), and social transfers to the neediest social groups, can protect human capital formation in all communities in periods of pandemics.

Our study suggests that specific interventions to reduce healthcare financing vulnerability (HFV) in the general population might not work. We have shown that HFVI is driven by five key factors:

(i) not having health insurance; (ii) lack of formal employment; (iii) lack of cash or liquidity; (iv) not having medical benefits at place of work; and (v) high catastrophic health expenditures, irrespective of ability to pay because these healthcare outlays can drive persons above the poverty line to extreme poverty (Ogwang & Mwabu, 2025), or push those below the poverty to destitution. Looking at (i) to (v) above, it is clear that HFV in a pandemic context is not just a lack of money, employment, or insurance. The general performance of the economy also matters, as the state of the economy determines whether or not the type of healthcare people want is available. Moreover, the needed healthcare might be available, but the national health system might not deliver it efficiently or equitably. Thus, in addition to specific interventions implied by the components of HFVI, policies to improve the performance of the whole economy are needed in crisis times. Equally important are governance institutions that can be relied upon to make the best possible social decisions in a pandemic context. Furthermore, critical in such a context is a mass communication technology to facilitate population-wide sharing of vital information about pandemics in real time.