Effects of health insurance on out-of-pocket payments: a coarsened exact matching analysis using Kenya’s 2022 demographic and health survey

Abstract

Objectives

Out-of-pocket payments remain a major source of healthcare financing in Kenya, yet it is unclear whether health insurance coverage is associated with lower out-of-pocket spending due to non-random enrolment. This study uses coarsened exact matching (CEM) to compare out-of-pocket costs (OOPCs) between insured and uninsured individuals.

Design, setting and participants

This study used cross-sectional data from the 2022 Kenya Demographic and Health Survey which covered 80 780 household members. Participants with missing or incomplete information on health insurance status and OOPC were excluded from the study.

Outcome measures and analysis

The main outcome variable was OOPC, computed from a summation of outpatient and inpatient costs less amounts paid by health insurance. CEM was applied to assess the influence of health insurance coverage on inpatient, outpatient and total healthcare costs. Gross healthcare expenditures (before insurance reimbursement) and net out-of-pocket payments (after insurance reimbursement) were analysed separately to distinguish healthcare utilisation from financial protection, with analyses conducted for different insurance types, including the now-defunct National Hospital Insurance Fund (NHIF), private insurance and community-based health insurance (CBHI).

Results

In the matched sample, insurance coverage was associated with higher gross inpatient, outpatient and total healthcare expenditures but significantly lower net OOPC for outpatient care (average treatment effect on the treated (ATET) = Kenya Shillings (KES) -5175; 95% CI −8755 to −1596), and total care (ATET = −KES6680; 95% CI −10 486 to −2874). The magnitude of financial protection varied by insurance type; private insurance and NHIF were associated with lower net out-of-pocket payments, while CBHI was associated with higher gross and net inpatient OOPC (ATET = KES12889; 95% CI 408 to 25370).

Conclusions

Among the matched insured population, these findings suggest that expanding insurance coverage alone may not be sufficient to achieve financial protection. Differences observed across schemes point to the potential importance of benefit design, reimbursement adequacy and risk pooling, particularly for schemes serving informal and low-income populations.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Coarsened exact matching improves comparability between insured and uninsured individuals by balancing observed covariates.

• The use of nationally representative data supports population-level inference.

• Separate analysis of gross and net out-of-pocket costs distinguishes healthcare utilisation from financial protection effects.

• Residual confounding may persist due to unobserved factors (eg, illness severity, care-seeking behaviour, quality of care).

Introduction

The attainment of universal health coverage (UHC) is a global priority. UHC aims to ensure that everyone has access to good-quality health services based on need without suffering financial hardship or impoverishment.¹ However, the WHO warns that countries are off-track for attaining UHC by the 2030 timeline as health service coverage has stagnated since 2015 while there has been an increase in the share of the population facing a financial catastrophe due to out-of-pocket costs (OOPC) incurred at the point of service use.² To move towards UHC, countries must expand population coverage with adequate prepayment financing mechanisms and reduce the share of the population that pays for healthcare services directly at the point of use. Similarly, countries must expand the range of health services provided to the population.³

Health financing arrangements that predominantly use prepayment arrangements rather than OOPC at point-of-service use are imperative for ensuring financial risk protection.^4–6 The main categories for organising prepayment financing in healthcare include a tax-funded approach and contributory health insurance schemes which can be in the form of social health insurance, private compulsory health insurance and voluntary health insurance (such as voluntary community-based health insurance (CBHI) and private voluntary health insurance).^{5 7} Contributory benefit entitlements characterise health insurance schemes. That is, making premium contributions is the provision for enjoying benefits in a given health insurance scheme and thus a determinant of access to care.⁵

Many low-and middle-income countries (LMICs), including Kenya, are reviewing their health financing systems to meet the principles of UHC. One financing mechanism, which has dominated UHC reforms in these countries is the development of contributory health insurance schemes. In theory, health insurance should protect individuals from incurring catastrophic costs at the time of illness, thereby promoting access to healthcare, particularly in settings where the government subsidises premiums for the poorest population.⁸ Nevertheless, evidence suggests that health insurance does not guarantee financial risk protection.^9–11 In addition, despite the consensus that OOPC is an inequitable mechanism for financing health systems, many LMICs continue to rely on OOPC to fund their health systems.^{4 6 12–16}

Kenya has committed to achieving UHC by 2030.¹⁷ The country’s healthcare system is pluralistic, with service provision provided by both public and private healthcare facilities in almost equal measure. The public healthcare delivery system is organised into four tiers, namely community, primary care, county referral and national referral.¹⁸ The country has a mixed health financing system that is financed through public, private and donor resources. According to the latest national health accounts, public sources (47.6%), donors (19.1%) and OOPC (26.6%) contributed to the country’s total health expenditure in the financial year 2018/2019.¹⁹ Table 1 summarises key health financing indicators in Kenya.

Table 1. Selected health financing indicators in Kenya.

Indicator	2001/2002	2005/2006	2008/2009	2013/2014	2015/2016	2016/2017	2017/2018	2018/2019
Percentage of THE financed by public sources	29.6%	29.3%	28.8%	33.5%	37.0%	46.0%	44.5%	47.6%
Percentage of THE financed by donors	16.4%	31.0%	34.5%	24.7%	23.4%	21.2%	22.3%	19.1%
Percentage of THE financed by private sources	54.0%	39.3%	36.7%	40.6%	39.6%	7.5%	8.1%	6.7%
Percentage of THE financed by out-of-pocket payments	N/A	N/A	25.1%	26.6%	26.1%	25.3%	25.2%	26.6%
Government health expenditure as a percentage of total government expenditure	7.9%	5.1%	4.8%	6.1%	6.7%	10.3%	10.7%	11.7%
THE per capita (USD)	51.2	59.5	66.3	77.4	78.6	97.4	99.9	105.81
THE as a percentage of GDP	5.1%	4.7%	5.4%	6.8%	5.2%	5.5%	5.3%	5.6%
Health insurance coverage	9.7%	N/A	10.0%	17.1%	19.0%	19.0%	19.0%	19.0%

Source: 2013/2014, 2015/2016 and 2018/2019 Kenya Health Accounts.

GDP, gross domestic product; THE, total health expenditure; USD, US dollars.

Public sources of revenue are collected by a mix of direct and indirect taxes, and premium contributions to the defunct National Hospital Insurance Fund (NHIF), now Social Health Insurance Fund (SHIF). The NHIF was a state corporation that was established in 1966 to provide mandatory health insurance to formal sector employees,^20–22 and its mandate later expanded to cover informal sector workers in 1998.²¹ Health insurance coverage in Kenya remains low, with only 26% of the population insured. This comprises 23% covered by the NHIF, 5% by private insurers and less than 1% by CBHI schemes.²³ The low health insurance coverage is perhaps not surprising given that Kenya, like most LMICs, has a disproportionately high informal sector population (83%) compared with the formal sector (17%).^{24 25} Among the NHIF reforms in the past decade, membership was made mandatory for all adults with the aim of increasing coverage of the informal sector.²⁶

As Kenya continues to search for a UHC solution, a major concern is whether having health insurance reduces OOP and protects households from incurring catastrophic health expenditures (CHE) and being impoverished. Although several studies have examined health insurance and financial protection in Kenya, existing evidence is mixed and often relies on descriptive or regression-based approaches^{9 12 21 27 28} that do not adequately account for the non-random nature of insurance enrolment.²⁹ Individuals who enrol in insurance differ systematically from those who do not in terms of socioeconomic status (SES), health status and healthcare-seeking behaviour,^{27 30} which complicates attribution of differences in out-of-pocket spending to insurance coverage alone. To our knowledge, there remains limited nationally representative evidence on the association between insurance coverage and out-of-pocket payments after accounting for observed confounding due to observed covariates.

Against this backdrop, using a causal inference framework, this study examines the influence of health insurance coverage, overall and by insurance type, on OOPC in Kenya using nationally representative data from the 2022 Demographic and Health Survey (DHS). To improve comparability between insured and uninsured individuals and reduce bias arising from observed non-random enrolment, the analysis applies coarsened exact matching (CEM), allowing for a more rigorous assessment of differences in OOPC associated with insurance coverage.

Methods

Data

This study used Kenya’s 2022 Demographic and Health Survey (KDHS), a nationally representative survey.²³ The KDHS employed a two-stage stratified sampling design. In the first stage, 1692 clusters were selected from the Kenya Household Master Sample Frame using the Equal Probability Selection Method. A household listing within these clusters provided the sampling frame for the second stage, where 25 households per cluster were randomly selected, totalling 42 022 sampled households. Out of these, 37 911 households were interviewed, with 19 747 completing the full questionnaire and 18 164 completing the short version. Insurance coverage data were collected through the full questionnaire, covering 80 780 household members. The survey also captured information on inpatient and outpatient out-of-pocket healthcare expenditures.

Study population

The study population comprised all individuals included in the 2022 KDHS for whom information was available on health insurance status, healthcare expenditures and covariates used in the matching procedure. Observations without suitable matches were excluded from the matched sample. Further details of the matching procedure are reported below.

Study variables

Health insurance coverage (treatment variable)

Information on the health insurance status of all household members was collected. For this study, insurance status was defined at the individual level and coded as a binary variable where the uninsured individuals were coded as ‘0’, while those with health insurance were coded as ‘1’. Additionally, data on the type of health insurance were categorised as follows: NHIF (3), private/commercial insurance (2), community-based insurance (1) and others (0). Dummy variables were created for each insurance type, with ‘no insurance’ serving as the reference category.

Out-of-pocket healthcare expenditure (outcome variables)

Three outcome measures were examined: inpatient OOPCs, outpatient OOPC and total OOPC. Inpatient costs were derived from household member reports of expenditures incurred during any overnight hospital admission in the 12 months preceding the survey. Reported amounts included all payments made for consultations, diagnostic tests and medicines, regardless of payment source (cash, insurance, in-kind or other). Two measures were constructed: (1) Gross inpatient costs, representing total reported expenditures before insurance reimbursement, and (2) Net inpatient OOPC, calculated by deducting the amount reimbursed by insurance from total reported expenditures.

Outpatient costs were derived from household member expenditures on care received from healthcare providers, pharmacies or traditional healers without an overnight stay in the past month. These costs were similarly calculated as gross outpatient OOPC and net outpatient OOPC after subtracting insurance reimbursements, and were annualised to ensure comparability with inpatient costs. Total OOPC, the third indicator, was computed by summing inpatient and outpatient costs for each household using STATA’s rowtotal function. If either inpatient or outpatient costs were missing, total OOPC was calculated using the non-missing component. Total OOPC was coded as missing only when both inpatient and outpatient cost components were missing. It should be noted that missing insurance payment amounts resulted in missing net OOPCs but did not affect analyses of gross expenditures.

Matching variables

Matching variables are used in matching estimator analyses to improve comparability between treatment (insured) and control (uninsured) groups by reducing confounding due to measured covariates. This study selected covariates based on their documented associations with both health insurance coverage and out-of-pocket healthcare costs, as identified in the literature.^{27 28 31–36} The chosen covariates included the age, sex, education level, marital status, employment status and chronic disease status of the household head (HH), dependency ratio (ratio of dependents (children and elderly) to working-age individuals in a household), SES and place of residence. Although insurance status was measured at the individual level, these household-level characteristics were included to account for shared economic and demographic conditions that influence both insurance enrolment decisions and healthcare spending. Thus, household-level variables were assigned to each individual within the household, consistent with standard practice in DHS analyses. All variables were used in their original DHS scaling, with categorical variables entered as factor variables and continuous variables (eg, age of the HH and dependency ratio) coarsened into substantively meaningful categories for matching. More details on these matching variables are provided in the results.

Conceptual framework

Figure 1 presents a directed acyclic graph that illustrates the hypothesised relationship between health insurance coverage and OOPC. Health insurance is considered the exposure, OOPC the outcome and healthcare utilisation a mediator through which part of the effect of insurance may operate. Insurance may reduce OOPC directly through financial protection and indirectly by influencing the use of healthcare services.

Figure 1. A directed acyclic graph (DAG) explaining the potential relationship between health insurance and OOPC.

Demographic and socioeconomic characteristics were treated as potential confounders because they may influence health insurance enrolment, healthcare utilisation and OOPC. The framework also acknowledges the possibility of residual confounding from unmeasured factors, such as health-seeking behaviour, illness severity, health awareness and risk preferences. Accordingly, the estimated outputs should be interpreted as conditional on the observed covariates included in the analysis.

Matching estimator analysis

The primary estimand of interest was the average treatment effect on the treated (ATET), defined as the difference in out-of-pocket healthcare expenditures that insured individuals would experience under observed insurance coverage compared with the counterfactual scenario of not being insured. Estimation of the ATET requires constructing a comparison group of uninsured individuals who are comparable to insured individuals with respect to observed confounding characteristics.

CEM was selected as the adjustment method to improve comparability between insured and uninsured individuals by improving imbalance in observed covariates while restricting the analysis to the region of common support.^37–40 Individuals without suitable matches were excluded from the matched sample; consequently, the estimated ATET applies only to insured individuals who could be matched to comparable uninsured individuals rather than to the full KDHS population.

Each matching variable was coarsened into substantively meaningful categories prior to matching. For example, age was grouped into standard adult age bands, education into attainment levels, SES into DHS wealth quintiles, and dependency ratio into low, medium and high categories. These cut points were chosen to reflect commonly used classifications in the literature^{27 28 31–36} and to balance bias reduction against sample retention. Sensitivity checks using alternative coarsening specifications for key variables (age and dependency ratio) were also conducted using finer and coarser cut points. Matching was performed separately for individuals with any insurance and for each insurance type (NHIF, private insurance and CBHI), with uninsured individuals serving as the comparison group in each case. Observations without suitable matches were excluded from the matched sample.

Covariate balance was assessed using both the multivariate L₁ imbalance statistic and standardised mean differences (SMDs) for individual covariates before and after matching. The L₁ statistic ranges from 0 to 1, with lower values indicating better overall balance across the joint distribution of covariates.^{37 41} SMDs were used to quantify the magnitude of imbalance for individual covariates independent of sample size, with absolute values below 0.10 considered indicative of acceptable balance. Prematching SMDs were calculated using the unmatched sample, while postmatching SMDs were calculated using the CEM weights to account for the weighting structure induced by the matching procedure.

Following matching and the assessment of covariate balance, the estimand, ATETs were estimated for inpatient, outpatient and total OOPCs both before and after deducting insurance reimbursements. This approach distinguishes differences in healthcare utilisation and spending patterns from the financial protection effect of insurance on net out-of-pocket payments. It should also be noted that the estimated ATETs apply to the subset of insured individuals who could be successfully matched to comparable uninsured individuals based on the selected covariates. Individuals whose characteristics fell outside the region of common support and for whom no suitable counterpart could be identified were excluded from the matched analysis. ATETs were estimated using weighted ordinary least squares regression with CEM weights and adjusting for the matching covariates. No measures of healthcare utilisation, service use or care-seeking behaviour were included in either the matching procedure or the postmatching regression because these variables were considered potential mediators of the relationship between insurance and OOPCs. All analyses were conducted using Stata V.17.

Ethical approval

All participants provided informed consent prior to data collection, and all procedures adhered to established ethical standards, including the principles of the Declaration of Helsinki.

Patient and public involvement

None.

Results

Descriptive statistics

Mean healthcare costs were higher prior to insurance reimbursement (table 2). Before accounting for insurance reimbursement, mean inpatient costs were Kenya shillings (KES)29 872, outpatient costs were KES22 674 and total healthcare costs were KES26 497. After deducting amounts paid by health insurance, mean out-of-pocket payments declined across all categories to KES21 161 for inpatient care, KES18 690 for outpatient care and KES20 824 for total healthcare expenditures.

Table 2. Summary of OOPC before and after health insurance reimbursements.

	Before deducting amounts paid by health insurance	After deducting amounts paid by health insurance
	Mean (95% CI)	Mean (95% CI)
Inpatient OOPC	29 871.73 (25 632.59 to 34 110.89)	21 161.44 (17 496.81 to 24 826.07)
Outpatient OOPC	22 674.27 (19 817.4 to 25 531.14)	18 689.55 (16 576.96 to 20 802.15)
Total OOPC	26 496.53 (23 782.27 to 29 210.79)	20 823.85 (18 806.54 to 22 841.16)

Amount in 2022 Kenya shillings (KES).

OOPC, out-of-pocket costs.

Characteristics of insured and uninsured participants before and after matching are presented in table 3. In the unmatched sample (n=80 350), insured (n=18 507) and uninsured participants (n=61 843) differed significantly across nearly all sociodemographic and household characteristics, including SES, place of residence, relationship to the HH, education, employment, sex, marital status, chronic disease status and dependency ratio (all p<0.001). Insured participants were disproportionately concentrated among wealthier households, male-headed households and those with higher educational attainment and employment, indicating substantial baseline imbalance between the two groups.

Table 3. Characteristics of insured and uninsured participants before and after matching.

	Before matching		After matching
	Insured % (95% CI)	Not insured % (95% CI)	Insured % (95% CI)	Not insured % (95% CI)
Individual-level insurance coverage	26.47 (25.38 to 27.60)	73.53 (72.40 to 74.62)	27.33 (26.22 to 28.48)	72.67 (71.52 to 73.78)
Mean age of HH (years)	44.53 (43.94 to 45.13)	46.25 (5.88 to 46.62)	44.66 (44.05 to 45.26)	46.84 (46.48 to 47.21)
Type of health insurance (individual level)
NHIF	22.96 (22.01 to 23.92)	_	23.71 (22.73 to 24.68)	_
Private insurance	5.21 (4.33 to 6.09)	_	5.34 (4.42 to 6.26)	_
CBHI	0.78 (0.61 to 0.95)	_	0.83 (0.65 to 1.01)	_
Household socioeconomic status
Poorest	0.98 (0.84 to 1.13)	18.78 (17.75 to 19.86)	1.03 (0.88 to 1.19)	18.06 (17.04 to 19.12)
Poorer	2.43 (2.17 to 2.72)	17.64 (16.80 to 18.51)	2.55 (2.27 to 2.86)	17.32 (16.49 to 18.19)
Middle	4.32 (4.01 to 4.65)	15.45 (14.68 to 16.26)	4.56 (4.23 to 4.92)	15.42 (14.64 to 16.24)
Richer	7.19 (6.64 to 7.78)	13.29 (12.41 to 14.24)	7.40 (6.83 to 8.02)	13.32 (12.41 to 14.28)
Richest	11.56 (10.45 to 12.78)	8.36 (7.58 to 9.21)	11.79 (10.65 to 13.04)	8.54 (7.76 to 9.40)
Place of residence
Urban	13.39 (12.39 to 14.46)	19.98 (19.04 to 20.96)	13.89 (12.87 to 14.99)	18.91 (17.97 to 19.88)
Rural	13.09 (12.42 to 13.78)	53.54 (52.32 to 54.76)	13.44 (12.76 to 14.15)	53.76 (52.51 to 55.00)
Relationship to HH
Head	9.55 (9.16 to 9.94)	16.54 (16.15 to 16.93)	9.25 (8.87 to 9.63)	14.74 (14.36 to 15.12)
Spouse	4.43 (4.20 to 4.66)	8.34 (8.09 to 8.60)	4.71 (4.47 to 4.96)	8.81 (8.54 to 9.09)
Children	10.58 (10.02 to 11.17)	34.66 (33.91 to 35.41)	11.40 (10.80 to 12.03)	36.28 (35.650 to 37.08)
Others	1.92 (1.72 to 2.13)	13.99 (13.53 to 14.47)	1.98 (1.78 to 2.20)	12.84 (12.36 to 13.33)
Education of HH
No education	1.53 (1.33 to 1.77)	9.35 (8.81 to 9.93)	1.47 (1.26 to 1.71)	8.18 (7.67 to 8.73)
Primary/incomplete secondary education	11.27 (10.64 to 11.93)	37.22 (35.99 to 38.46)	12.24 (11.53 to 12.98)	36.07 (34.81 to 37.35)
Secondary or higher	23.79 (22.52 to 25.12)	16.83 (16.00 to 17.70)	24.85 (23.51 to 26.24)	17.19 (16.29 to 18.13)
Marital status of HH
Single	1.68 (1.45 to 1.93)	3.42 (3.11 to 3.76)	8.32 (7.69 to 9.00)	26.99 (26.34 to 27.65)
Married	19.56 (18.77 to 20.38)	50.35 (49.20 to 51.51)	20.39 (19.56 to 21.25)	33.49 (32.53 to 34.47)
Widowed/separated/divorced	4.22 (3.88 to 4.60)	19.67 (18.89 to 20.48)	2.62 (2.39 to 2.86)	8.18 (7.80 to 8.58)
Sex of HH
Female	6.95 (6.49 to 7.44)	24.91 (24.05 to 25.79)	6.76 (6.31 to 7.25)	23.66 (22.80 to 24.54)
Male	19.52 (18.57 to 20.51)	48.61 (47.49 to 49.74)	20.57 (19.58 to 21.60)	49.01 (47.87 to 50.14)
Employment status of HH
Unemployed	1.84 (1.50 to 2.26)	7.71 (6.97 to 8.52)	2.17 (1.71 to 2.75)	4.77 (4.14 to 5.49)
Employed	25.53 (24.26 to 26.72)	64.19 (62.85 to 65.52)	37.41 (35.80 to 39.05)	55.65 (53.92 to 57.36)
Chronic disease status of HH
No chronic illness	24.91 (23.84 to 26.01)	60.99 (59.65 to 62.31)	37.17 (35.82 to 38.53)	60.44 (59.07 to 61.79)
One chronic illness	1.73 (1.49 to 2.01)	2.15 (1.90 to 2.44)	1.39 (1.14 to 1.69)	1.01 (0.84 to 1.20)
More than two chronic illnesses	0.00	0.25 (1.60 to 0.39)	_	_
Dependency ratio*
No dependency*	4.93 (4.47 to 5.43)	7.95 (7.51 to 8.41)	4.81 (4.34 to 5.33)	6.75 (6.34 to 7.17)
Medium*	17.70 (16.81 to 18.62)	48.56 (47.46 to 49.67)	18.60 (17.69 to 19.55)	49.78 (48.64 to 50.92)
High*	3.88 (3.50 to 4.29)	16.99 (16.23 to 17.77)	3.92 (3.54 to 4.35)	16.14 (15.39 to 16.91)
N	18 507	61 843	16887	46701

All values are percentages with 95% CIs, except mean age, which is reported in years. Characteristics are shown for the full sample stratified by insurance status.

Bold values show population size before and after matching.

^*No dependency applies to households without dependents, while medium dependency includes those where each working-age individual supports fewer than two dependents. High dependency implies that each working-age person supports at least two dependents.

CBHI, community-based health insurance; HH, household head; NHIF, National Health Insurance Fund.

Following CEM, 16 887 insured and 46 701 uninsured individuals were retained in the matched sample (n=63 588), corresponding to approximately 79.14% of the original sample, while unmatched observations were excluded to enforce common support. The matched sample therefore represents the subset of individuals for whom comparable insured and uninsured counterparts exist based on observed characteristics. Postmatching, the marginal distributions of covariates were well balanced between groups, indicating substantial improvement in comparability. Within the matched sample, 27.3% of individuals were insured and 72.7% uninsured, with similar distributions across key characteristics such as employment status, chronic disease status and household headship.

Table 4 summarises mean inpatient, outpatient and total healthcare expenditures among insured and uninsured individuals in the matched sample, reported before and after deducting amounts paid by health insurance. Before deducting insurance payments, insured individuals incurred substantially higher mean healthcare expenditures than uninsured individuals across inpatient, outpatient and total care. Inpatient costs were particularly high among the insured (KES77 494) compared with the uninsured (KES12 289).

Table 4. Summary of OOPC in the matched group.

	Before deducting amounts paid by health insurance		After deducting amounts paid by health insurance
	Mean (95% CI)		Mean (95% CI)
	Insured	Not insured	Insured	Not insured
Inpatient OOPC	77 493.50 (56 433.91 to 98 553.08)	12 289.08 (10 101.57 to 14 476.59)	38 289.46 (23 242.46 to 53 336.47)	13 731 (11 327.01 to 16 134.99)
Outpatient OOPC	35 494.83 (26 479.97 to 44 509.69)	14 532.04 (12 818.29 to 16 245.79)	20 655.58 (16 321.98 to 24 989.17)	14 797.18 (13 044.46 to 16 549.91)
Total OOPC	51 281.39 (39 898.90 to 62 663.89)	15 122.48 (13 542.98 to 16 701.99)	27 299.77 (21 844.70 to 32 754.84)	15 668.79 (14 020.01 to 17 317.56)
N	17 758	56 036	17 758	56 036

Amount in 2022 Kenya shillings (KES).

OOPC, out-of-pocket costs.

Similar patterns were observed for outpatient and total expenditures. After deducting amounts paid by health insurance, mean out-of-pocket payments among insured individuals in the matched sample declined markedly across all care categories. Inpatient out-of-pocket payments among the insured decreased to KES38 289, while outpatient and total out-of-pocket payments declined to KES20 656 and KES27 300, respectively. In contrast, net out-of-pocket payments among uninsured individuals in the matched sample remained largely unchanged, reflecting the absence of insurance reimbursement. Despite these reductions, insured individuals continued to incur higher absolute out-of-pocket payments than uninsured individuals, particularly for inpatient care.

Validity of matching

Table 5 presents covariate balance before and after CEM using both the multivariate L₁ imbalance statistic and SMDs. Before matching, substantial imbalance was observed between insured and uninsured individuals, as reflected by a multivariate L₁ statistic of 0.554 and several large SMDs. The largest imbalances were observed for household SES (SMD=1.08), educational attainment of the HH (SMD=0.89), relationship to the HH (SMD=0.45), place of residence (SMD=0.43) and dependency category (SMD=0.31). These findings indicate considerable differences in observed socioeconomic and demographic characteristics between insured and uninsured individuals prior to matching.

Table 5. L₁ statistic and SMDs before and after CEM.

	Insured–not insured
	Before matching		After matching
	L1 (mean)	SMD	L1 (mean)	SMD
Household socioeconomic status	0.374 (1.238)	1.075	1.3e-14 (9.7e-14)	0.000
Place of residence	0.177 (−0.177)	0.429	7.0e-15 (1.5e-14)	0.000
Relationship to HH	0 (0)	0.445	1.1e-14 (6.5e-14)	0.000
Education of HH	0.409 (0.526)	0.888	7.6e-15 (9.3e-15)	0.000
Marital status of HH	0.092 (−0.092)	0.146	1.1e-14 (2.1e-15)	0.000
Sex of HH	0.027 (0.027)	0.141	1.7e-15 (−1.2e-15)	0.000
Age	0.062 (0.140)	0.050	1.2e-14 (−4.6e-14)	0.000
Employment status of HH	0.075 (0.075)	0.241	3.2e-15 (4.6e-15)	0.000
Chronic disease status of HH	0.042 (0.038)	0.111	4.5e-15 (4.7e-16)	0.000
Dependency ratio	0.085 (−0.128)	0.311	2.1e-14 (4.9e-14)	0.000
Multivariate L1	0.554		2.083e-14
N	80 350		63 588

CEM, coarsened exact matching; HH, household head; SMD, standardised mean difference.

Following CEM, covariate balance improved substantially. The multivariate L₁ statistic decreased from 0.554 before matching to 2.07 × 10⁻¹⁴ after matching. Similarly, all weighted SMDs were reduced to approximately zero, demonstrating that the matching procedure improved balance between insured and uninsured individuals on the covariates included in the matching process.

The reduction in sample size from 80 350 to 63 588 observations reflects the exclusion of individuals without suitable matches, enforcing common support.

Estimates of the effects of health insurance on OOPC

Table 6 reports the ATET from the CEM analysis, comparing insured individuals with matched uninsured individuals.

Table 6. Association between health insurance and OOPC in the matched cohort with CEM weights.

Outcome variables	Before deducting amounts paid by health insurance		After deducting amounts paid by health insurance
	ATET (95% CI)	P value	ATET (95% CI)	P value
Any insurance
Inpatient costs	34 057.85*** (20 144.85 to 47970.85)	0.000	−2166.28 (−12 215.30 to 7882.74)	0.672
Outpatient costs	6547.42*** (1919.58 to 11 175.26)	0.006	−5175.45*** (−8754.69 to −1596.22)	0.005
Total OOPCs	14 070.94*** (6509.03 to 21 632.85)	0.000	−6680.37*** (−10 486.33 to −2874.41)	0.001
NHIF
Inpatient costs	29 456.37*** (−5240.90 to 64 153.63)	0.000	−1939.82 (−12 829.48 to −8949.84)	0.727
Outpatient costs	5048.25** (769.45 to 9327.04)	0.021	−4183.70*** (−8066.81 to −300.60)	0.005
Total OOPCs	9778.79*** (1975.21 to 17 582.37)	0.000	−5673.40 (−9766.13 to −1580.66)	0.156
Private insurance
Inpatient costs	80 520.91*** (59 681.33 to 101360.50)	0.000	−12404.82 (−27 869.01 to 3059.36)	0.116
Outpatient costs	33 944.37*** (21 583.41 to 46 305.34)	0.000	−13 137.31*** (−22 300.95 to −3973.67)	0.005
Total OOPCs	34 032.40*** (20 996.75 to 47 068.06)	0.000	−16 318.72** (−26 636.25 to −6001.20)	0.002
CBHI
Inpatient costs	13 155.98** (2503.05 to 23808.92)	0.016	12 888.89** (408.00 to 25 369.60)	0.043
Outpatient costs	−5046.57 (−11 293.70 to 1200.56)	0.113	−1262.80 (−9477.49 to 6951.89)	0.358
Total OOPCs	-892.15 (−7455.30 to 5670.99)	0.590	2993.29 (−5850.33 to 11836.92)	0.411

ATET, average treatment effects on the treated; CBHI, community-based health insurance; CEM, coarsened exact matching; NHIF, National Health Insurance Fund; OOPC, out-of-pocket cost.

Any health insurance

Before deducting insurance payments, individuals with any health insurance were associated with higher outpatient costs (ATET = KES 6547; 95% CI 1920 to 11 175; p=0.006), higher inpatient costs (ATET = KES 34 058; 95% CI: 20 144 to 47 791; p=0.000) and higher total healthcare expenditures (ATET = KES 14 071; 95% CI 6509 to 21 633; p=0.000) compared with matched uninsured individuals. After deducting insurance payments, insurance coverage was associated with lower net out-of-pocket payments for outpatient (ATET = −KES 5175; 95% CI −8755 to −1596; p=0.005) and total costs (ATET = −KES 6680; 95% CI −10 486 to −2874; p=0.001). These patterns indicate that insured individuals tend to incur higher gross healthcare expenditures but lower net out-of-pocket payments relative to comparable uninsured individuals.

National Hospital Insurance Fund

Before deducting amounts paid by health insurance, NHIF coverage was associated with significantly higher gross inpatient costs (ATET = KES 29 456; 95% CI −5241 to 64 154; p=0.000), outpatient costs (ATET = KES 5048; 95% CI: 770 to 9327; p=0.021) and total healthcare costs (ATET = KES 9799; 95% CI: 1975 to 17 582; p=0.000) relative to matched uninsured individuals.

After deducting insurance payments, NHIF coverage was associated with lower net OOPC for outpatients (ATET = −KES 4184; 95% CI −8067 to −301; p=0.005). However, the differences in net inpatient and total OOPC were not statistically significant. These findings suggest that NHIF membership was associated with higher gross healthcare expenditures but provided statistically significant financial protection only for outpatient care.

Community-based health insurance

CBHI coverage was associated with higher gross inpatient costs compared with uninsured individuals (ATET = KES 13 156; 95% CI 2503 to 23 809; p=0.016), while gross outpatient costs were lower (ATET = -KES 5047; 95% CI −11 293 to 1201; p=0.113). After deducting insurance payments, CBHI remained associated with higher net inpatient out-of-pocket payments (ATET = KES 12 889; 95% CI 408 to 25 370; p=0.043), and no statistically significant associations were observed for outpatient or total net OOPCs (p=0.358 and p=0.411, respectively). These patterns suggest that CBHI membership is associated with higher inpatient expenditures that are not fully offset by insurance reimbursements, relative to comparable uninsured individuals. However, these findings should be interpreted with caution given the relatively small proportion of CBHI-covered individuals in the sample, which may contribute to potential instability in the estimates.

Private insurance

Private insurance was associated with substantially higher gross inpatient (ATET = KES 80 521; 95% CI 59 681 to 101 361; p=0.000), outpatient expenditures (ATET=KES 33 944; 95% CI 21 583 to 46 305; p=0.000) and total healthcare expenditures (ATET=KES 34 032; 95% CI 20 997 to 47 068; p=0.000). After deducting insurance payments, private insurance was associated with the lowest net OOPC among all insurance types, including outpatient costs (ATET = -KES 13 137; 95% CI −22 301 to −3974; p=0.005) and total costs (ATET = -KES 16 319; 95% CI −26 636 to −6001; p=0.002). Compared with other insurance schemes, private insurance was associated with the lowest net out-of-pocket payments. These patterns suggest that private insurance coverage is associated with higher gross healthcare expenditures alongside lower net out-of-pocket payments relative to comparable uninsured individuals.

Discussion

This study examined the influence of health insurance coverage and OOPC in Kenya using CEM to improve comparability between insured and uninsured individuals. Three main findings emerged. First, matched individuals with any health insurance were associated with higher gross inpatient, outpatient and total healthcare expenditures than comparable uninsured individuals; however, once insurance reimbursements were accounted for, insurance coverage was consistently associated with lower net out-of-pocket payments for outpatient and total care. Second, the magnitude of these associations varied across insurance types, with private insurance showing the strongest association with lower net OOPCs, followed by the NHIF. Third, CBHI was associated with higher gross and net inpatient OOPC, perhaps suggesting comparatively weaker financial protection relative to other schemes.

The finding that insurance coverage in the matched sample is associated with higher gross healthcare expenditures but lower net out-of-pocket payments is consistent with the expected role of insurance in facilitating access to healthcare while reducing direct financial burden at the point of use.^{15 42 43} Higher gross expenditures among insured individuals may reflect greater utilisation of services that may otherwise be deferred or forgone due to cost barriers, a finding documented in other LMICs.^{44 45} At the same time, lower net out-of-pocket payments after reimbursement indicate that insurance mitigates the financial burden of healthcare spending, aligning with the financial risk protection objective of UHC.¹ These dynamics are closely patterned by socioeconomic inequality and should be interpreted within that context. However, this pattern may also partly reflect residual differences between insured and uninsured individuals that could not be fully accounted for in the matching procedure. For example, individuals with greater underlying healthcare needs, more severe illness, stronger health-seeking preferences or better awareness of available healthcare services may be both more likely to enrol in health insurance and more likely to use healthcare. To the extent that these factors were not measured in the data set, some of the observed differences in healthcare expenditure may reflect residual confounding rather than the effect of insurance coverage alone.

Observed differences in healthcare spending likely reflect not only variation in utilisation alone but also differential access shaped by income, employment and place of residence.⁴⁶ In this context, a key finding is that matched insured individuals may, in some cases, incur higher absolute out-of-pocket payments than matched uninsured individuals, even after reimbursement, particularly for inpatient care. While this may partly reflect greater utilisation among the insured, it also points to the role of factors such as benefit design (including co-payments, ceilings and exclusions) and the tendency for insured individuals to access more intensive or higher-cost care. Importantly, lower observed healthcare expenditures among uninsured individuals (particularly those in lower-income households) should not be interpreted as evidence of lower need or superior financial protection, but rather as a likely reflection of foregone or unmet care due to inability to pay, resulting in reduced or absent utilisation.⁴⁷ Conversely, insured individuals may obtain more care than they would otherwise have accessed while still making some out-of-pocket payments.⁴⁷ As such, the associations observed in this study reflect dynamic interactions between insurance status, access, benefit design and financial capacity, rather than simple reductions in spending, and higher observed spending among the insured should not be interpreted solely as increased utilisation, nor lower spending among the uninsured as evidence of better financial protection.

However, the magnitude of financial protection varied by insurance scheme. Among matched insured individuals, NHIF and private insurance were associated with lower net out-of-pocket payments compared with being uninsured, but private insurance demonstrated a substantially stronger association with reduced net costs across outpatient and total care. This may likely reflect differences in benefit design, reimbursement ceilings, provider payment arrangements and coverage depth.^{15 36} In contrast, although matched NHIF was associated with lower net out-of-pocket payments, particularly for outpatient care, the magnitude of protection was more modest. This finding is consistent with previous assessments of NHIF that highlight limitations related to benefit caps, partial coverage and balance billing.^{9 21 27} Despite the relatively stronger associations observed for matched private insurance, coverage remains low (approximately 5%) and is concentrated among higher-income earners in urban areas, raising important equity considerations.

As a result, privately insured individuals are likely to consume more (and more expensive) healthcare and therefore realise larger absolute savings from insurance reimbursement. In contrast, poorer and uninsured individuals may be excluded from care altogether due to affordability constraints, resulting in lower observed expenditures but substantial unmet need. These dynamics suggest that differences in observed OOPC reflect both financial protection and unequal access to care. Additionally, private insurance policies often require co-payments, which can reduce their effectiveness in protecting against catastrophic and impoverishing health expenditures.^48–50 Integrating aspects of private insurance, such as efficient reimbursement mechanisms and broader benefit packages, into Kenya’s public health insurance could enhance financial protection and equity in healthcare access.⁴³

The results for matched CBHI differ substantially from those for matched NHIF and private insurance. CBHI coverage was associated with significantly higher gross and net inpatient OOPC compared with being uninsured. This pattern may reflect limited financial protection against high-cost hospital care, rather than differences in utilisation alone. This interpretation is consistent with concerns raised in previous research regarding the sustainability and effectiveness of CBHI schemes in LMICs.^{34 51} However, it is important to note that CBHI schemes in Kenya are heterogeneous, varying in benefit design, reimbursement arrangements and depth of coverage. The DHS data used in this study do not capture these scheme-level characteristics, and the CBHI category therefore reflects an aggregate of potentially diverse models. As such, the observed patterns should be interpreted as an average effect across schemes, rather than the performance of any specific CBHI programme.

Despite studies in Ethiopia and Rwanda demonstrating that well-structured CBHI models may reduce OOPC,^52–54 CBHI schemes are typically characterised by small risk pools, limited revenue bases and constrained benefit packages, which reduce their ability to absorb large inpatient costs.⁵⁵ As a result, reimbursement levels may be insufficient when members experience severe illness requiring hospital admission.

In addition, adverse selection may play a role, as individuals with higher expected healthcare needs may be more likely to enrol in CBHI schemes, particularly in the absence of mandatory enrolment.⁵⁶ The large magnitude of inpatient costs observed among CBHI members may reflect a combination of high-cost hospital episodes and limited reimbursement, rather than over-reporting of expenditures. Residual selection on unobserved illness severity, where individuals with greater healthcare needs are more likely to enrol in CBHI, may also contribute to this pattern. Individuals anticipating greater healthcare needs or requiring more frequent healthcare services may be more likely to enrol in CBHI schemes, particularly where enrolment is voluntary. Consequently, part of the higher inpatient expenditures observed among CBHI members may reflect differences in underlying healthcare need rather than differences in insurance performance alone. To improve CBHI’s financial protection role in Kenya, policymakers should consider integrating CBHI with the formal health insurance system, providing targeted subsidies to make CBHI more affordable, expanding benefit packages to include inpatient and outpatient care, and strengthening risk pooling mechanisms, initiatives which have shown success in Rwanda and Ethiopia.^{42 53 54 57} However, while the CBHI findings are useful, they should be viewed as indicative rather than definitive, especially given the small sample of the CBHI analyses and the lack of scheme-level detail in the data.

Together, these findings suggest that among insured individuals represented within the matched sample, insurance coverage alone may not be sufficient to ensure effective financial risk protection. The depth of coverage, adequacy of reimbursement and ability to cover high-cost inpatient services are critical determinants of whether insurance reduces out-of-pocket payments.^{1 5 6} Despite recent expansion, health insurance coverage in Kenya remains low at 27%, leaving 73% of the population uninsured.²³ Among insured individuals, NHIF accounted for most coverage (82%), followed by private insurance (15%) and CBHI (2%).²³ The results of this study indicate that uninsured individuals are associated with higher out-of-pocket payments, while the degree of financial protection among insured individuals varies across schemes.

As Kenya advances health financing reforms and transitions towards SHIF, these results highlight the need to move beyond nominal coverage expansion towards strengthening benefit design, provider payment mechanisms and risk pooling, particularly for schemes intended to protect poorer and more vulnerable populations.^{7 24} Although matched insurance coverage was generally associated with lower net out-of-pocket payments, the degree of financial protection varied substantially across schemes. The stronger protection observed among matched privately insured individuals suggests that comprehensive benefit packages, efficient reimbursement systems and greater depth of coverage are important features that could inform SHIF implementation.

The findings also underscore the need to address persistent barriers to enrolment among poorer and more vulnerable populations, including liquidity constraints, unaffordable contributions, irregular incomes and high attrition rates. Without adequate subsidies and effective mechanisms for enrolling informal-sector workers, those most vulnerable to CHE may remain uninsured, undermining both equity and risk pooling.

Furthermore, the relatively modest protection observed for NHIF and the substantially higher inpatient expenditures associated with CBHI suggest that insurance coverage alone does not guarantee effective financial protection. As SHIF is implemented, strengthening coverage for high-cost inpatient services, reducing co-payments and balance billing, ensuring timely provider reimbursement, and improving risk pooling will be essential for protecting households from healthcare-related financial hardship.

Addressing long-standing challenges such as affordability of contributions, delayed reimbursements and limited inpatient coverage^{9 24} will be essential if Kenya’s health insurance efforts are to deliver meaningful financial protection and advance UHC. Lessons from Kenya’s NHIF experience and from other LMICs^{58 59} suggest that targeted subsidies, mandatory enrolment, strategic purchasing and efficient claims processing will be central to ensuring that SHIF provides equitable and sustained protection against out-of-pocket healthcare spending.

Study strengths

This study has several strengths. First, the present study used the latest data (2022) from KDHS, a nationally representative survey.²³ This provided a broader sampling frame for this analysis in as much as the estimated treatment effects apply only to the subset of insured individuals who were successfully matched to comparable uninsured individuals. Second, we employed CEM, an advanced econometric methodology, to improve comparability between insured and uninsured individuals by reducing imbalance in observed pretreatment covariates.^{37 40} While this approach strengthens internal validity, it does not eliminate the possibility of residual confounding. Third, the analysis covered multiple health insurance types including the NHIF, private insurance and CBHI, thus offering better insights into the varied effects of these schemes on OOPC. Lastly, separate analyses for gross and net OOPC helped distinguish healthcare utilisation from financial protection effects.

Study limitations

The findings of this study should be interpreted considering the following limitations. First, the reliance on self-reported data for OOPC introduces the potential for recall bias, as respondents may struggle to accurately recall healthcare expenditures, especially for services received within the past year. Second, despite the use of matching technique, unmeasured confounders, such as health-seeking behaviour, severity of illness, prior healthcare utilisation and the quality of healthcare services accessed, may still influence the observed outcomes. Thus, these findings should be interpreted as associations conditional on observed covariates rather than causal effects. Third, the relatively small sample sizes for CBHI and private insurance groups may limit the precision and generalisability of findings for these schemes. This reduction in sample size reflects limited overlap between insured and uninsured individuals with comparable observed characteristics. In addition, because CEM excludes observations outside the region of common support, the estimated ATET across entire findings should not be interpreted as applying to all insured individuals in Kenya or the entire KDHS population.

Finally, the cross-sectional nature of the KDHS data limits the ability to establish temporal ordering between health insurance coverage and healthcare expenditure. Because exposure and outcome were measured at a single point in time, it is not possible to determine whether observed differences in expenditure were caused by insurance coverage or whether underlying differences in healthcare need and healthcare-seeking behaviour influenced both insurance enrolment and expenditure patterns. Longitudinal studies would be valuable for strengthening causal inference and assessing the longer-term effects of health insurance on financial risk protection.

Conclusion

These findings suggest that expanding insurance coverage alone may not be sufficient to ensure financial risk protection. As Kenya transitions to the SHIF, reforms could prioritise strengthening and broadening benefit packages, ensuring adequate reimbursement for high-cost inpatient care and improving risk pooling, particularly for populations with low and unstable incomes, to better align coverage with financial protection.

Data availability statement

Data are available in a public, open access repository.