We used the publicly available Primary Health Care Costing, Analysis, and Planning (PHC-CAP) Tool (20) to facilitate the calculation of actual and normative costs. The PHC-CAP Tool is an activity-based costing tool based in Microsoft Excel. Actual costs are calculated using data gathered from a sample of health facilities which are extrapolated to the entire universe of facilities in the area of interest. Normative costs represent the resources needed to provide the services with reasonable quality and efficiency (19). Normative costs are estimated using standard treatment protocols (STPs) for each health service and intervention (21) in the PHC package, with the estimated populations in need for each service based on the population size (reported by sub-counties) and expected service need given estimated rate of disease incidence and prevalence as well as utilization rates for promotional and preventive services. The difference between actual and normative costs represents the resource gap. Aligned with the World Health Organization’s (WHO) PHC Measurement Framework and Indicators (22), the PHC-CAP Tool incorporates key metrics including total cost, cost per capita, cost per service and program, cost per input, numbers of inpatient and outpatient services per clinical staff, and average daily service output per clinical staff. In addition to this application in Kenya, the PHC-CAP Tool has been used in five other countries including Ethiopia (23) and Nigeria (24).

We used multi-stage sampling to select the health facilities included in the study. In a first stage, we selected six out of 47 counties in consultation with the Ministry of Health (MOH; Supplementary 3). Of the six counties, three (Nyeri, Kisumu, and Isiolo) were selected due to their participation in the Afya Care UHC pilot which entitles registered households to free services in level 4 and level 5 public hospitals (25), two (Kajiado and Kilifi) were chosen given their geographic diversity (Rift Valley and coastal, respectively), and one (Makueni) was selected for its health financing scheme “Makueni Care” which provides free care for registered users at public hospitals (26). In a second stage, we selected seven sub-counties based on the following criteria: (a) presence of a referral hospital and (b) outpatient per capita utilization rate was within the median range for that sub-county based on 2018–2019 data in the Kenya Health Information System (KHIS). One sub-county was selected in each county except in Nyeri where two were selected.

Within each sub-county, data were collected from each sub-county hospital and a sample of health centers, dispensaries, and community health units (CHUs) attached to each facility. The sample of health centers was based on total outpatient utilization – i.e., with preference for those with higher outpatient utilization. Dispensaries for which the total outpatient utilization figures were closest to the 25% quartile, 50% quartile, and 75% quartile of the overall distribution were selected. No sampling was conducted for CHUs, as their services and costs were captured at the facility level. All facilities sampled were public except for the faith-based Mary Immaculate Hospital which was the only referral hospital in Kieni West. The sampled facilities comprised 38 dispensaries, 26 health centers, seven sub-county hospitals, and associated CHUs (Table 1). Although our aim was to sample (on average) 10 health facilities in each of the seven sub-counties, there was some variation. In some sub-counties (e.g., Kilifi where we sampled six), the number of sampled facilities fell below this aim due to time and travel constraints while in others (Kisumu where we sampled 16), we were able to sample more facilities.

Actual costs, irrespective of funding source, reported by health facilities (July 2018–June 2019) included recurrent health facility operating costs (e.g., electricity, water), human resources (i.e., staff salaries of clinical and non-clinical personnel), and other recurrent costs. Drug and medical supply costs were reported through the Kenya Medical Supplies Authority (KEMSA) database. Data on facility expenditures were reported directly from facilities using checklists and semi-structured interviews; catchment population sizes and budgets were reported through county and sub-county annual work plans. Data on capital costs and above-facility costs (e.g., national and county support and supervision) were excluded from this analysis. The study was conducted from the public sector perspective and did not set out to capture patient costs, including out-of-pocket expenditures on drugs and medical supplies. Data collection was conducted from February to June 2020. The currency exchange rate used was 1 US dollar (USD) equal to 101.99 Kenyan schilling (Ksh) to allow for international comparison (27).

We collected data on the number of inpatient and outpatient services for each facility and sub-county from the KHIS system. Only services reported from July 2018 to June 2019 were included in this analysis to ensure consistency with the cost analysis. Service data were reported monthly and categorized according to KHIS service classifications. For example, the KHIS reported the number of family planning attendances by facility but did not report the disaggregated quantity of individuals receiving oral contraceptives, condoms (male or female), or injectable contraceptives. Although most sub-county hospitals and health centers reported numbers of inpatient cases (for persons younger and older than 5 years old), only five of seven sub-county hospitals reported data on the case mix of inpatient services. Moreover, PHC services provided by CHUs were not separated from facility-based services.

We calculated total annual PHC costs for each facility in the sample. Costs of clinical labor, drugs, medical supplies, and other indirect costs were summed and allocated to outpatient visits and inpatient days based on weighted service volumes in dispensaries, health centers, and sub-county hospitals using WHO-CHOICE data to create the relative cost of an outpatient visit compared to an inpatient day (28). This same WHO-CHOICE ratio of inpatient to outpatient costs was also used to estimate weighted daily caseloads of clinical staff: a ratio of 432:126 in primary hospitals, 432:111 in health centers, and 432:90 in dispensaries (28). In the absence of reliable time use data by service type, this method (i.e., using WHO-CHOICE ratios) was particularly useful at higher level facilities (health centers and hospitals) which provided both inpatient and outpatient services, allowing us to allocate costs according to their estimated resource use.

To estimate the actual total cost per capita of PHC in the seven sub-counties beyond the sampled facilities to the universe of facilities (i.e., extrapolation), we used the following three-step approach (29). First, for each of the facility types (dispensary, health center, and hospital), we compared the total number of outpatient services in our sample with the number of outpatient services in the universe of facilities (also obtained from the KHIS system), and inflated our sample costs in proportion. This follows the assumption that our sample is representative of the sub-counties, and Kenya in general, in terms of outpatient service unit cost. Second, we added the estimates for dispensaries, health centers, and hospitals within a given sub-county, and Kenya in general. Lastly, the actual cost per capita was calculated by dividing the total actual cost by the reported total sub-county or Kenyan population.

Normative costs were estimated based on STPs which were developed for clinical services included in the PHC service package, excluding those not provided at health facilities.¹ STPs were developed for approximately 24% (73 of 309) services in the PHC package, corresponding to more than 80% of all services provided. Each outpatient service from the Kenya PHC package of services was consolidated and grouped by service-related disease category (i.e., matched) according to the corresponding KHIS service grouping classification. Each service group category received proportional weight in its contribution to the overall cost per KHIS groupings based on previous frequency of use. Normative costs per KHIS grouping were calculated and reported for each level of the PHC network (hospital, health center, and dispensary).

For each STP, information was gathered on the following: total drug, diagnostic, and laboratory reagent requirements per service episode and the corresponding unit cost for each; average number of annual encounters per service episode; total human-resource requirements, based on the number of minutes required per service by human resource cadre and the corresponding salary cost; population in need, computed based on the population of each sub-county and disease incidence and prevalence rates for each service sourced from either national data or from the 2019 Global Burden of Disease (GBD) Kenya dataset (30). Each STP was developed and reviewed by members of an expert panel comprised of MOH staff representing various departments² with a subsequent review conducted by a primary care physician.

We assumed that 30% of clinical health worker time was dedicated to necessary non-clinical activities (e.g., administration and reporting, training, and time spent between consultations) which we considered a conservative estimate for the amount of non-clinical work necessary to provide quality PHC services for the entirety of the catchment population. To account for indirect costs not considered in the STPs, such as non-clinical personnel, utilities and other operational expenses, an overhead rate was calculated based on our sampled facilities, and added to the normative human resource, drug, and medical supply costs. The overhead rate for each facility level was calculated by dividing total indirect cost by the sum of clinical labor, drug and medical supply costs, and indirect costs. Using this approach, we projected the normative costs for public sector PHC facilities to deliver high-quality care for an expected service volume which reflects all needed services among catchment populations in the specified geographic area. This projection, aiming at UHC, assumes that all (100%) needed PHC services are met. Although aspirational, this scenario is indicative of the maximum service coverage expected.

We conducted one-way sensitivity analyses to understand the impact of various factors on the actual and normative costs per capita and displayed them in tornado diagrams. For both the actual and normative cost per capita, we conducted the same variable adjustments and also estimated best- and worst-case scenarios by varying all of these factors together. We varied clinical labor costs by ±20% given the potential uncertainty of facility reported salaries, the reported problem of ghost workers on government payroll (and potential to curb such costs) (18), and future plans to remunerate CHVs. We varied drug costs by ±30% and supply costs by ±20% to account for any data inaccuracies in KEMSA. We varied indirect costs by ±30% since the indirect costs incorporated in the normative estimates were derived from our actual costs. We also varied population by ±5% given the uncertainty of the figures reported in annual workplans.

Health facility staffing patterns within the sample varied by sub-county and by facility level (Table 2). Sub-county hospitals had an average (mean) of 100 total staff (ranging from 43 in Garbatulla to 175 in Kibwezi West) with an average of 65 clinical staff (range 21–143). Among the 26 sampled health centers, there were an average of 36 total staff (range 2–109), with an average of 30 clinical staff (range 2–92) with some health centers reporting no non-clinical staff support.³ Among the 38 sampled dispensaries, there were an average of five total staff (range 3–20) and three clinical staff (range 1–17).

Among the sampled facilities, hospitals accounted for the highest per facility volume of outpatient services and inpatient bed days, with fewer inpatient services delivered at the health center levels. Among the different facility types, there was a considerable range of total outpatient services delivered annually – e.g., in Garbatulla, one dispensary only provided 2,355 services while a dispensary in Kajiado East provided more than 43,000 services. In terms of the total volume of outpatient services delivered within the sampled sub-counties (Table 3), dispensaries provided the majority of services (46.23%), followed by health centers (29.74%), and hospitals (24.03%). This contrasted slightly with national-level service data reported from the KHIS for dispensaries (40.8%), health centers (25.6%), and hospitals (33.7%).

The daily caseload of clinical staff (also shown in Table 2) highlights the variations in staffing patterns and service output. On average, clinical staff provided 3.0 outpatient services per day at hospitals, 3.7 at health centers, and 16.3 services per day at dispensaries. When weighing both outpatient and inpatient services according to WHO-CHOICE ratios, hospitals accounted for more daily caseloads per clinical staff. Due to data availability, a more in depth analysis of outpatient case mix was not possible as it would have required detailed data from health facility registries. Moreover, only six sampled hospitals reported inpatient data and only five hospitals recorded inpatient case mix, therefore a complete analysis of inpatient service delivery was not possible.

Among the facilities sampled, average annual costs of delivering PHC services varied considerably across facility types and the sub-counties (Table 4): dispensaries (US$ 12,794 – US$ 237,486); health centers (US$ 30,235 – US$ 1.5 M); and hospitals (US$ 498,674 – US$ 5.5 M). Clinical labor costs represented the largest cost driver among all facility types, followed by drugs, indirect costs, and medical supplies (Figure 1).

Based on the extrapolation of sampled facility costs to the universe of facilities, the total delivery cost in the seven sub-counties ranged from $US 1.3 million in Garbatulla to US$ 9.3 million in Kajiado East (Table 5). Based on reported sub-county populations, this corresponds to a per capita cost ranging from US$ 9.30 in Ganze to US$ 47.2 in Mukurweini. The total annual PHC delivery costs in Kenya were estimated at $US 1.36 billion (US$ 26.6 per capita) when the sample was extrapolated to all dispensaries, health centers, and sub-county hospitals using KHIS service data. Hospitals made up the vast majority of costs (US$ 948 million).

Table 6 compares the annual actual per capita PHC cost in the seven sub-counties with the normative cost per capita of delivering the Kenya PHC package at universal coverage. The national actual cost per capita (US$ 26.6) was substantially lower than the normative cost per capita (US$ 41.7), which corresponds to a sizeable resource gap of US$ 15.1 per person and the need to increase investment by 1.6 times its current level. The actual cost per capita was lower than the normative cost per capita in all sub-counties with the exception of Mukurweini which suggests there is no financial resource gap in that specific county to achieving universal coverage of the Kenya PHC package.

The sensitivity analyses are presented as tornado diagrams in Figures 1, 2. When clinical labor costs, drug costs, supplies costs, indirect costs, and population size were changed simultaneously to create best- and worst-case scenarios, the resulting actual cost per capita ranged from US$ 18.8 to US$ 35.3 (Figure 2). Meanwhile, the normative cost per capita ranged from US$ 30.3 to US$ 54.3 (Figure 3). Normative costs, which were estimated with other assumptions of necessary non-clinical time, are available in Supplementary 4.

This study offers valuable evidence on sub-national cost variations and resource requirements to guide the implementation of the GOK’s PHC reforms and resource mobilization efforts. Assuming that the entire Kenyan population would access PHC services in public facilities, our study demonstrates the need for increasing investment by 1.6 times its current level nationally. This suggests that PHC is currently underfunded and would require a substantial increase in financial resources to expand PHC service provision.

Nevertheless, closing the PHC resource gap may also be possible by addressing existing inefficiencies within the health system. For example, our study suggests that there is considerable underutilization of labor and low caseloads among clinical staff in sampled facilities. Low staff utilization among clinical personnel could be due to numerous factors, including, but not limited to staff absenteeism and low demand for services due to geographic or financial barriers or perceived quality of care. This variance in productive efficiency could represent available staff time that could be optimized if the PHC system experienced an increase in service utilization.

When examining labor utilization (i.e., daily services provided per clinical worker), many low-volume facilities spent more financial resources relative to higher-volume facilities, suggesting an inefficiency in labor use. Nevertheless, there are some facilities that report both higher utilization and have relatively lower costs, suggesting that clinical staff spend less time with patients. Our suppositions on inefficiencies within the health facilities are consistent with previous work. A number of previous studies on outpatient services in Kenya have highlighted the inefficiencies that exist in health facilities and that there is variation across facilities (32, 33). Previous work has also highlighted important variation in efficiency across county health systems in Kenya with some counties being quite efficient and others being very inefficient (34).

Furthermore, our study highlights the considerable variation in staffing patterns among health facilities. Although the difference in staffing patterns could be necessary to meet the needs of its catchment populations (especially for facilities in more urban areas requiring more personnel), it could also indicate that health facilities lack clear guidelines for staffing norms. The maldistribution of clinical staff could be due a lack of clear guidance for staffing norms or, as suggested by Nyawira et al. (18), the potentially high prevalence of ghost workers, staff absenteeism, and general over-staffing at higher levels of care (18).

Our findings also suggest there is opportunity to improve the overall efficiency of care by managing referrals through gatekeeping and better informing patients of the referral process (35), as described in the PCN guidelines (3). For example, in Makueni County, patients are required to receive prior authorization from the referral facility before seeking services while providers are on a fee-for-service basis to incentivize high-priority PHC services (36). Similarly, in Nyeri County, the pilot UHC program assigned patients to the nearest health facility where they would initially seek health services so as to reduce overcrowding at sub-country hospitals. The NHIF could also potentially limit reimbursement for certain services provided at hospitals, and strategic purchasing mechanisms (e.g., performance-based financing) could be leveraged to incentivize service provision at specific PHC system levels. Based on KHIS data, hospitals provided more than one-third of the PHC services nationally, indicating there is potential to increase care-seeking at health centers, dispensaries, and CHUs. Although challenging to determine the ideal ratio of service provision across facility levels, there were some categories of services (e.g., diarrhea, malaria, pneumonia) which we would expect to be treated at lower facility levels or CHUs; however, a considerable share are still provided at hospitals.⁴ Unfortunately, the KHIS does not specify the severity of each case, the number of cases with co-morbidities, nor upward and downward referrals.

This study did have a number of limitations. The study assumed that the sampled facilities were representative of the PHC network in the seven sub-counties which may limit the expansion of cost results to the universe of facilities and at a national level (for Kenya). Nevertheless, our sampling approach goes beyond many costing studies which commonly use convenience sampling (37). The study also relied on service utilization data from KHIS which in the past has been reportedly known to have quality and completeness issues. Due to aggregation of KHIS categories, it is possible that hospitals also provided secondary services which are not included in the Kenya PHC package, which would have resulted in an overestimate of actual PHC costs. Although national policy prohibits level 2 facilities (dispensaries) from providing inpatient services, our sample indicated a small number of inpatient services were provided. Normative costs were based on standard treatment guidelines and expert panel opinion which are subject to bias. Moreover, individual STPs did not account for potential time savings due to integrated service provision (e.g., simultaneous diagnosis and treatment of multiple conditions). Moreover, Kenya-specific incidence and prevalence estimates for STPs were unavailable for certain conditions and had to be approximated using the best available evidence.

Additional analysis on staffing mix, task shifting (upstream and downstream), and service outputs (including case mix, quality of service provision, and levels of efficiency) is necessary to inform opportunities for improving PHC efficiency. Further analysis should also be conducted to determine the level of and reasons for bypass, as well as the impact of existing gatekeeping mechanisms. Our analysis also underscores the necessity of improving the nominal classification of health facilities and ensuring adequate human resources are equipped to provide explicit services within the PHC package. Further insight is required into facilities with low service volume per clinical staff (e.g., low demand for services, absenteeism), highlighting the need to closely examine labor patterns and slack capacity to better understand if this is due to idle staff time or absenteeism. Improving the capture of CHV-level data to assess their level of service provision is also critical. Future cost analyses would benefit from more detailed service data within the KHIS (e.g., severity of diagnosis and inpatient case mix data), as well as routinely reported cost and human resource data (number of staff per cadre per facility level). This would reduce the need for sub-county and facility level data collection.