Technical and scale efficiency in health service production in Kenya: subnational analysis of 47 county governments

Abstract

Background

Kenya adopted a decentralized health system to improve access to care and accelerate progress toward universal health coverage (UHC). County governments now manage most health service delivery; however, the extent to which resources are used efficiently remains unclear.

Methods

We compiled county-level data on health inputs, outputs, and contextual factors for 2022. Overall technical efficiency (OTE), pure technical efficiency (PTE), and scale efficiency (SE) were estimated using the single-step Simar and Wilson Data Envelopment Analysis (DEA) approach. Contextual determinants of efficiency were examined using truncated regression.

Results

Counties demonstrated moderate efficiency for child survival (OTE = 0.74; 95% CI: 0.70–0.78), indicating potential to improve outcomes by about 26% using existing resources. PTE averaged 0.85, suggesting that 15% of inefficiency stemmed from managerial factors, while SE averaged 0.87, reflecting a 13% loss attributable to suboptimal scale. For childhood immunization, the average OTE was 0.83 (95% CI: 0.81–0.87), implying that coverage could increase by 17% without additional inputs; uniformly high PTE points to strong managerial performance in immunization programs. Maternal survival showed the greatest inefficiency (OTE = 0.51; 95% CI: 0.48–0.55), with counties achieving only half of their potential. Efficiency for healthy life expectancy (HALE) was moderate (OTE = 0.83; 95% CI: 0.81–0.85), indicating scope for a 17% improvement without new resources. Substantial heterogeneity was observed across counties for all outputs.

Conclusion

In 2022, maternal health services exhibited markedly lower efficiency than child health and immunization programs. The severe inefficiency in maternal service delivery, compared to child health and immunization programs, highlights an urgent need for targeted managerial strengthening in low-performing counties such as Kisii and Nyeri. Strengthening managerial capacity and optimizing scale of operations could yield major health gains without additional resources.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13690-026-01969-1.

Text box 1. Contributions to the literature
• Introduces a flexible approach to integrating multi‑source data for evaluating subnational health system performance in LMICs.
• Reframes policy dialogue toward efficient use of existing resources amid fiscal constraints.
• Provides a scalable framework for benchmarking county health performance.
• Highlights routine efficiency measurement as an accountability tool for planning and resource allocation.

Background

Kenya’s successive governments have long viewed investments in health as foundational to national development. [1,2] Policies such as Vision 2030 emphasize expansion of health service delivery. [3] In 2010, the country adopted a two-tier health governance structure that devolved service delivery to 47 autonomous counties. [4] These counties were formed based largely on historical and political considerations rather than geographic or economic criteria. [5].

Under this governance model, the national government retained responsibility for setting standards, formulating policy, coordinating national programs, mobilizing resources, and providing technical support to counties. [6] The central government also allocates health resources to counties according to population size and historical inequities. [7] Counties, in turn, plan, organize, and deliver health services for their residents—an essential function for achieving Kenya’s universal health coverage (UHC) target by 2030. [8,9].

In response, counties have invested in expanding infrastructure, recruiting health workers, and procuring medicines and equipment. [10,11] These investments assume that county health systems use resources efficiently. However, studies in Africa and other regions suggest that technical and allocative inefficiencies remain widespread. [12] For example, a district-level study in Zambia showed that the health system operated at less than 62% of optimal capacity. [13] Studies from India and China similarly reported large geographical disparities in efficiency across maternal and non-communicable disease services. [14,15] In addition, multiple analyses have documented low technical efficiency in various HIV interventions in Nigeria, South Africa and Rwanda. [16].

In Kenya, health sector investments have yielded substantial gains. [17] National life expectancy rose by 5·4 years between 1990 and 2016, and maternal mortality declined, although trends varied widely across counties. [17] Under-five mortality showed similar heterogeneity, with some counties achieving rapid improvements and others showing little progress. Despite gains, Kenya’s improvements lag those of peer countries with similar development levels. [17,18]

Several studies have examined Kenya’s health system performance, but most focused on specific programs or facilities rather than the overall system. [19–21] Others assessed efficiency using service volumes, which assume uniform quality and overlook potential adverse effects. These gaps restrict understanding of how effectively county systems convert available resources into improved population health. To address these limitations, this study employs health outcomes -- rather than activity measures -- as outputs for efficiency analysis to assess factors associated with county-level performance. [21–23] Specifically, this study seeks to answer the following questions:

1. How efficiently do Kenyan counties transform available health resources into key health outcomes?

2. To what extent do managerial and scale factors drive variations in county performance?

3. What contextual characteristics, if any, are associated with differences in efficiency?

To our knowledge, this is the first study in Kenya to assess technical and scale efficiency using a comprehensive, county-level perspective.

Methodology

Efficiency analysis

Health efficiency analyses typically distinguish between technical, allocative, and scale efficiency, and between input- and output-oriented approaches. [24] Because input prices needed to estimate allocative efficiency were unavailable, we only considered technical and scale efficiencies. We adopted an output-oriented model, reflecting the reality that county health systems face fixed inputs in the short run and can primarily influence how efficiently those inputs are used to generate health outcomes.

We measured technical efficiency using Data Envelopment Analysis (DEA), a non-parametric linear programming method that accommodates multiple inputs and outputs. [25] Appendix 1 provides the algebraic specification of the DEA model. We applied the widely used Charnes-Cooper-Rhodes (CCR) [26] and the Banker-Charnes-Cooper (BCC) [27] formulations. The CCR model assumes constant returns to scale and yields overall technical efficiency (OTE), while the BCC model assumes variable returns to scale and produces pure technical efficiency (PTE). Scale efficiency (SE) was calculated as the ratio of OTE to PE. [28, 29]

OTE reflects how close a county is to the best-performing frontier overall. For example, an OTE of 0.70 indicates that a county could theoretically achieve the same health outcomes using 30% fewer inputs, or conversely, generate 30% more output with current inputs. PTE isolates managerial performance; a PTE of 0.85 means that 15% of potential output is lost due to inefficiencies unrelated to scale. SE indicates whether a county is operating at an optimal scale of production. An SE score of 0.80 suggests that 20% of inefficiency is attributable to counties being either too large or too small relative to the scale-efficient frontier. [28, 29] All three measures were estimated for each of Kenya’s 47 counties.

Because Kenya’s health system is structured around vertical programs, we assessed efficiency using four distinct health outputs. [30] To contextualize county performance, we also grouped counties according to the former provincial boundaries. Although Nairobi was previously designated as its own province, we classified it under the Central region given its proximity to the region.

We estimated efficiency using the single-step Simar and Wilson’s (S-W) bootstrap approach, which generates bias-corrected efficiency scores and simultaneously evaluates determinants of efficiency. [31] DEA does not require specifying a functional form for the production frontier, making it suitable for heterogeneous county system. [32] A truncated regression embedded within the S-W framework was used to assess the influence of contextual variables. Bias-corrected confidence intervals were obtained through 10,000 bootstrap replications. [33, 34]

Efficiency analysis requires three categories of variables: inputs, outputs, and contextual factors. Inputs included population-adjusted health workforce (doctors, clinical officers, nurses, and midwives) and the development health budget. Outputs included child immunization coverage, health-adjusted life expectancy (HALE), under-five year survival, and maternal survival. The latter two were computed as the reciprocals of under-five mortality rate (U5MR) and maternal mortality rate (MMR) respectively for each county. Although counties control key health inputs, they operate in diverse environments that may affect efficiency. [35, 36] We therefore included women’s education, access to clean water, and access to electricity as contextual variables.

Data and variable description

Selection of contextual variables was guided by data availability and prior evidence identifying them as important determinants of health. Education, particularly among women of reproductive age, is strongly associated with improved health outcomes through pathways like improved health-seeking behavior, decision-making autonomy, and fertility preferences. [3, 13] Access to clean water is a well-established determinant of child survival, [1,12,13] while access to electricity serves as a proxy for reliable household energy. In settings without electricity, reliance on polluting fuels increases indoor air pollution, an important health risk. [12,13]

County-level data for 2022 were compiled from multiple sources. Health workforce, development health budget, and population data were obtained from the Kenya National Bureau of Statistics (KNBS). [37, 38] Measures of childhood immunization coverage, women education, access to water and electricity were drawn from the Kenya Demographic and Health Survey of 2022, also conducted by the KNBS. HALE, U5MR and MMR were sourced from subnational estimates for Kenya, produced by the Institute for Health Metrics and Evaluation (IHME) at the University of Washington. [39] Each health system output was considered separately using the same inputs and contextual variables to provide insight into specific dimensions of the health system performance. All variables were log-transformed to reduce skewness and stabilize variance, thereby improving the robustness of DEA and regression estimates.

Data analysis and management procedures

Data management, descriptive analyses, and visualizations were conducted using Stata version 16.1 (StataCorp, College Station, TX). [40] Simar–Wilson single-step analyses were implemented in R using the rDEA package, with 10,000 bootstrap replications to obtain bias-corrected efficiency estimates. This number of replications exceeds the minimum recommended by Simar and Wilson. [41].

Results

Descriptive statistics

Health system inputs and outputs varied substantially across Kenya’s 47 counties. Development health budgets ranged widely (10.40–196.12 million KES per 100,000 population), as did health workforce density (6.75–54.74 workers per 10,000 population). Outputs—including child immunization, under-five mortality, maternal mortality, and HALE—also showed marked heterogeneity, underscoring differences in baseline health system capacity and epidemiological context (Table 1).

Table 1.

Descriptive Statistics of Variables

Indicator	Unit of Measurement	Mean	SD	Min	Max
Health System Inputs
Development Health Budget	KES in millions per 100,000 population	53.92	39.41	10.40	196.12
Health Workforce Density	Health worker per 10,000 population	16.40	8.00	6.75	54.74
Health System Outputs
Under Five Mortality Rate	Deaths per 1,000 live births	42.53	11.70	15.00	73.00
Immunization Coverage	Percentage (%) of under 5 population	78.24	15.86	22.90	95.90
Maternal Mortality Rate	Deaths per 100,000 live births	264.69	168.38	64.66	855.95
Healthy Life Expectancy	Years	55.88	2.71	48.97	59.39
Environmental Factors
Percentage of Households with Access to Electricity	Percentage of Households	47.66	20.21	16.30	94.50
Percentage of Women Educated Above Secondary School	Proportion of women aged 15–49	50.14	16.73	13.60	77.40
Percentage of Households with Access to Clean Water Source	Percentage of Households	61.82	16.60	22.80	89.70

Overall efficiency, pure technical efficiency and scale efficiency

Child survival

Figure 1 shows the distribution of overall technical efficiency (OTE), pure technical efficiency (PTE), and scale efficiency (SE) using child survival as the output. About 30% of counties (14 of 47) achieved an OTE of 0.80 or above, with lower performance concentrated in parts of Central, Coast, Rift Valley, and Nyanza. In contrast, PTE was relatively strong, with 35 counties (75%) scoring ≥ 0.80, although pockets of low managerial efficiency remained in North Rift Valley, Nyanza, and Western regions. SE scores were high for most counties, with isolated weaknesses in Eastern and Rift Valley. Appendix 2 provides full county rankings.

Fig. 1.

Efficiency of child survival

Childhood immunization

As shown in Fig. 2, childhood immunization was the most efficient service area. Two-thirds of counties (32 of 47) had OTE values ≥ 0.80, though lower performers clustered in Coast, Eastern, Rift Valley, and Western. All counties exceeded PTE of 0.80, reflecting generally strong managerial capacity. SE was more variable, with lower values in Central and Eastern counties. Appendix 3 lists full county-level results.

Fig. 2.

Efficiency in childhood immunization coverage

Maternal survival

Maternal survival exhibited the greatest inefficiency (Fig. 3). Only three counties achieved an OTE ≥ 0.80, and just ten reached a PTE ≥ 0.80. Low performance was widespread across Central and Eastern regions, with no consistent spatial pattern. The national average OTE was 0.51, indicating that maternal health services operate at roughly half of their potential across the country. SE averaged 0.82, suggesting that while scale contributes to inefficiency, managerial factors account for the larger share. Appendix 4 provides detailed rankings.

Fig. 3.

Efficiency of maternal survival

Healthy life expectancy (HALE)

For HALE (Fig. 4), 28 counties (≈ 60%) recorded OTE ≥ 0.80, with lower scores concentrated in Central, Coast, Rift Valley, and Nyanza. SE followed a similar geographic pattern, with 31 counties (≈ 66%) scoring ≥ 0.80. PTE remained uniformly high nationwide. Appendix 5 shows the complete county rankings.

Fig. 4.

Efficiency of healthy life expectancy

Provincial patterns and correlations

Provincial-level rankings (Appendix 6) differed by outcome: Eastern led in child survival, Western in childhood immunization, and Rift Valley in maternal survival. OTE values across outputs were strongly correlated (Fig. 5), including child immunization with child survival (r = 0.74), child immunization with HALE (r = 0.96), and child survival with HALE (r = 0.81), indicating common underlying performance drivers across service areas.

Fig. 5.

Correlation of technical efficiency scores

Contextual factors

In truncated regression models, none of the contextual variables -- women’s education, electricity access, or clean water access -- showed statistically significant associations with efficiency scores across any of the four outputs (Table 2). This indicates that unmeasured factors, such as managerial capacity or local governance, may play a larger role in shaping county-level efficiency.

Table 2.

Truncated Regression Results

	Child Survival	Immunization	Maternal Survival	*HALE
Intercept	1.72 (0.90, 2.60)	1.70 (1.04, 2.38)	-7.21 (-17.79, 12.76)	1.54 (0.84, 2.28)
Education	0.05 (-0.15 0.23)	− 0.07 (-0.22, 0.06)	− 0.63 (-3.30, 2.36)	− 0.07 (-0.23, 0.09)
Water access	− 0.08 (-0.31, 0.13)	− 0.05 (-0.23, 0.11)	0.94 (-3.21, 3.49)	− 0.02 (-0.20, 0.16)
Electricity Access	− 0.05 (-0.20, 0.11)	− 0.01(-0.13, 0.12)	1.81 (-1.17, 3.81)	0.00 (-0.14, 0.13)

*HALE Health-adjusted life expectancy. 95% confidence intervals in parentheses

Discussion

Efficient use of resources is central to achieving UHC, [42] especially in the current environment of declining external health financing. Decision makers in resource-constrained countries like Kenya must increasingly identify sustainable approaches to financing and organizing care. [43, 44] Recent reductions in global health assistance underscore the importance of ensuring that existing resources are used efficiently. [45] These pressures coincide with growing demands on health systems and other competing government priorities. [46].

Our findings indicate substantial inefficiencies across Kenyan county health systems, with the largest gaps observed in maternal health services, where average performance reached only 51% of potential output. For under-five survival, counties could reduce mortality by an estimated 26% using existing inputs. Notably, half of the lowest-performing counties for child survival were in Nyanza and Coastal regions -- areas characterized by high malaria burden and limited access to basic amenities -- suggesting continued influence of local epidemiological and socioeconomic conditions. [1, 2, 47] Yet the presence of several low-performing counties in the relatively wealthier Central Province indicates that additional factors, including managerial or organizational constraints, may also play an important role.

Managerial and organizational capacity emerged as important contributors to inefficiency. Improving PTE alone could reduce U5MR by approximately 15%, with substantial county-level variation. For example, Migori could achieve a 27% improvement in child survival outcomes by attaining managerial performance levels observed in Mandera, the best-performing county. Given the autonomy afforded to counties under Kenya’s devolved system, such disparities may reflect differences in stewardship, decision-making quality, and local accountability mechanisms.

Scale inefficiency also contributed significantly to suboptimal performance. If all counties operated child health programs at optimal scale, U5MR could fall by an additional 13%. Counties showed both decreasing returns to scale (indicating systems that are too large) and increasing returns to scale (systems that are too small). Most operated under increasing returns to scale, suggesting that expanding service delivery capacity could yield efficiency gains. [25–27, 48] These findings highlight the importance of aligning county program size with population needs and service demands.

Efficiency levels were highest for childhood immunization, where counties could increase coverage by 17% without new resources. High PTE across counties suggests that managerial practices developed through Kenya’s long-standing immunization program remain strong. [17, 38, 49] However, SE findings indicate that additional gains could be achieved by reorganizing service delivery models to address population dispersion and reach underserved areas. Innovations such as expanded community health strategies and digital tools for tracking immunization could further enhance performance. [50, 51].

Maternal health showed the greatest inefficiency, with counties operating at only half of their potential. PTE estimates indicate that improved managerial performance alone could reduce MMR by nearly 39%, while SE improvements could contribute an additional 18%. Counties such as Nyeri and Kisii have substantial room for improvement in managerial practices, whereas counties like Marsabit and Samburu stand to benefit more from optimizing scale.

Maternal survival efficiency is far lower than childhood immunization because maternal care requires a complex, multi-step service network -- skilled staff, emergency obstetric capacity, reliable referral and transport, blood supply, and continuous readiness -- where failures at any point greatly affect outcomes. [7–9, 51] Immunization, by contrast, relies on standardized, low-complexity contacts delivered through strong outreach platforms supported by Kenya’s long-standing childhood immunization infrastructure, resulting in uniformly high managerial efficiency. [2–8] Maternal services also involve higher financial and time costs for families, which delay timely care, and socio-cultural factors -- such as care-seeking norms, birth preferences, and concerns about quality or respectful care -- further influence facility delivery. [2–8, 19]

Kenya could improve HALE by approximately 17% without new inputs. Much of this potential gain stems from scale adjustments, mirroring patterns observed for child survival and immunization—reflecting the contribution of these conditions to overall population health. Strong correlations among OTE scores across outputs further support this relationship and suggest system-wide opportunities for efficiency improvement.

Child survival clusters mirror disease ecology and basic infrastructure gaps, especially in high-malaria or low-amenity settings. Immunization efficiency clusters reflect programmatic strengths -- cold-chain reliability, outreach capacity, and logistical execution -- consistent with high PTE nationwide. Maternal survival shows no stable spatial pattern because constraints vary widely: some counties struggle mainly with managerial inefficiencies (low PTE), others with scale misalignment (low SE), and some with both. Historical investment patterns, governance quality, accountability mechanisms, and terrain or population dispersion further influence these shifting patterns. [17, 38, 52, 53–55]

Kenya’s devolved governance structure places responsibility for resource allocation and service delivery at the county level. As a result, national progress toward UHC will increasingly hinge on county-level capacity to plan, manage, and monitor health programs effectively. Addressing both technical and scale inefficiencies is particularly important amid fiscal constraints and reductions in external support. [56, 57] Strengthening cross-county learning platforms, communities of practice, and data-driven decision-making could accelerate improvements. Counties could benefit from sharing practical managerial strategies, reorganizing service delivery scale, and adopting context-specific models such as mobile outreach or integrated service platforms. [58–61]

Evidence from other African settings reinforces the managerial and scale interpretation. Zambia’s subnational DEA documented considerable slack and scale effects in child health services, echoing our findings on returns to organizational improvement and right-sizing. Ghana hospital DEA analyses and multi-country outpatient HIV efficiency (including Nigeria, South Africa, Rwanda) similarly attribute much of inefficiency to management practices and facility/system scale, not just resource scarcity. Region-wide UHC efficiency work in sub-Saharan Africa also highlights the salience of contextual development indicators and program design nuances over raw input levels.

Limitations

This study has several limitations that should be considered when interpreting the findings. First, DEA measures the efficiency of each decision-making unit relative to the best performers within the sample rather than against an external or absolute benchmark. [25–27] As a result, a county deemed efficient in this analysis is only efficient relative to other Kenyan counties and may still underperform compared with theoretical optima or health systems in other settings.

Second, although we relied on the most recent nationally available data sources, these datasets were originally collected for different purposes and vary in quality. We were unable to include potentially important environmental variables -- such as governance quality, disease burden, or geographic characteristics -- that are likely to influence health system performance. Similarly, private health expenditures could not be incorporated due to lack of county-level data. This omission may have led to overestimation of efficiency in counties where private spending constitutes a substantial share of total health resources. To help mitigate this concern, we included childhood immunization -- largely funded through public sources -- as one of the outputs, and its patterns were consistent with our broader findings.

Third, inputs and outputs were measured for the same year (2022). Because health investments often have lagged effects, efficiency may be mis-estimated in counties that recently made substantial changes in staffing, infrastructure, or financing. Although major input shocks were unlikely in the preceding years, this constraint remains important, as DEA assumes contemporaneous production. [25–27] Using cross-sectional data therefore may obscure real improvements or declines that had not yet translated into measurable outcomes. Future analyses would benefit from multi-year, longitudinal input–output series to more accurately capture dynamic efficiency.

Fourth, the absence of input price data prevented estimation of allocative efficiency. [21, 24–28] Our analysis therefore focuses exclusively on technical and scale efficiency. We attempted to mitigate this limitation through careful data curation and application of robust non-parametric methods.

Finally, DEA is sensitive to outliers. [13, 21, 25–27] To examine the stability of our results, we conducted sensitivity analyses by iteratively removing one county at a time and refitting the model (47 models in total). Comparisons of root-mean-square error (RMSE) values and pairwise correlations across these models showed < 2% deviation in most cases, with strong correlations overall, suggesting that outliers are unlikely to have biased our efficiency estimates.

Conclusion

In this analysis, we assessed the technical and scale efficiency of Kenya’s county health systems while accounting for contextual factors that help explain performance differences. Several policy-relevant lessons emerge for Kenya and other countries with similar socioeconomic characteristics. First, substantial heterogeneity across counties, driven by differences in geographic size, population distribution, and service needs, suggests that decisionmakers should tailor service delivery models to achieve economies of scale more effectively. The national Ministry of Health should work with high-IRS counties -- such as Turkana, Elgeyo-Marakwet, Marsabit, and other sparsely populated counties -- to pilot integrated service delivery models or mobile outreach clinics to expand population reach. These models would consolidate service platforms, reduce duplicative fixed costs, and help counties move closer to the scale-efficient frontier.

Second, efficiency patterns show regional clustering, with provinces trading positions depending on the health output examined. This likely reflects varying epidemiological burdens– such as malaria, HIV/AIDS, and non-communicable diseases– as well as demographic and socioeconomic factors that shape service utilization and outcomes. [17, 49, 50] Efforts to improve health system performance should therefore be adapted to the specific epidemiological and demographic contexts of each county. [50–53] Counties with the lowest PTE for maternal survival, especially Kisii and Nyeri, should be prioritized for targeted, data-driven interventions focused on strengthening the emergency obstetric care (EmONC) network, referral reliability, staffing coverage, and quality assurance mechanisms.

Third, effective health programming must address structural determinants of health, including socioeconomic development, education, and both financial and physical barriers to care. These considerations are particularly relevant to Kenya’s recent reforms under the Social Health Authority (SHA), which aim to expand financial protection and reduce access barriers. [55].

Fourth, successful planning, resource mobilization, and implementation require balancing the autonomy of Kenya’s devolved county governments with the need for efficient and equitable service delivery. [57] Addressing both technical and scale inefficiencies is especially timely given the tightening fiscal environment and reductions in donor funding across the African region. [58, 59].

Finally, creating structured opportunities for shared learning could strengthen managerial and organizational capacity. Communities of practice or evidence-sharing networks focused on priority domains of health system management would enable counties to exchange practical strategies, benchmark performance, and enhance accountability toward national health goals. [60–62].

Overall, health system leaders must systematically monitor and link resources to population-level outcomes, using comprehensive and routine data to guide planning. Our findings highlight the importance of addressing inefficiencies -- particularly inappropriate scale and suboptimal managerial capacity -- to accelerate progress toward Kenya’s 2030 UHC agenda. Lessons from Kenya’s experience may also inform reforms in other countries with comparable health system structures and constraints.

Authors’ contributions

TA conceptualized the study. NR, DB and JT collated the data and did preliminary analysis. TA, YK, WO, MS and LW did data analysis. TA and YK wrote the first draft and WO, LW, RW, TO, AL, UA, and MS did the detailed review.

Funding

This study was not funded.

Data availability

The main data supporting the conclusions of this article are available on request and with written permission from KNBS. The data from IHME are publicly available from their website.

Declarations

Ethics approval and consent to participate

Permission to conduct the study was granted by the Ministry of Health, Kenya and followed established national guidelines. Our study did not involve human participants and only used available secondary data at the county level. It was determined by the Ethics Committee at Africa Institute for Health Policy that no formal ethics approval was required in this case.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.