Cost-Effectiveness of Emergency Obstetric Care in rural Kenya: Comparing Ambulance transfer and Self-referral

Obstetric complications are dicult to predict and may require referral, expedited by ambulance use. We conducted a cost-effectiveness analysis comparing ambulance transfers and self-referrals in obstetric emergencies in a predominantly rural setting in Kenya. Methods A retrospective cross-sectional cost-effectiveness analysis using a healthcare system perspective was conducted of parturient women transferred by ambulance to a higher level hospital compared with self-referrals between January to June 2019. Direct costs needed for ambulance, self-referral and clinical care were calculated. Every woman admitted with a pregnancy-related complication was assessed using the adapted sub-Saharan African Maternal Near Miss (MNM) criteria. Each referred woman was categorized as: ‘necessary referral’ meaning that they were managed for either MNM or potentially life-threatening complications (PLTC) and ‘unnecessary referral’ meaning those with no obstetric complications. Incremental cost effectiveness ratio (ICER) for referral was considered attractive or very attractive interventions when costs per life years gained (LYG) were below $150 and $30, respectively.

rate of 4.3 [15]. Rate of skilled birth attendance was 52.2% in 2014 [15]. The total number of 145 health facilities consists of one public referral hospital (Longisa county referral hospital, LCRH), two faith-based hospitals, ve sub-county hospitals, 23 health centers and 114 dispensaries. All public health services are nanced by the county government of Bomet. In LCRH, services include comprehensive emergency obstetric and newborn care (CEmONC) and 24-hour ambulance services free-of-charge to transfer women from their villages or health centers in case of obstetric complications. The faith-based and sub-county hospitals also provide CEmONC services and have access to the same ambulance services. The county government of Bomet owns two ambulances and hires four from Kenya Red Cross Society; three of the hired ambulances are located in in LCRH of which are our economic evaluation assessment was based. Each ambulance operates independently, covering distances between LCRH and health centers ranging from 2 to 48 km.
This was a facility-based cross-sectional study involving a retrospective chart review among women who had been admitted with obstetric complications brought in either by ambulance or as self-referrals between January, 1st, to June, 30th, 2019 in LCRH, Bomet, Kenya. Eligibility for the study was independent of gestational age: all women who arrived in LCRH with pregnancy related complications were included. Women who developed complications more than 42 days after termination of pregnancy were excluded. Every woman admitted with a pregnancy related complication was assessed using the adapted sub-Saharan African Maternal Near Miss (MNM) criteria [16]. Each referred woman was categorized as: 'necessary referrals' meaning that they were managed for either MNM or potentially lifethreatening complications (PLTC) and 'unnecessary referrals' meaning those with no obstetric complications. A necessary referral was assumed to be a woman referred from a lower to a higher level of care, meaning LCRH, where either MNM or PLTC cases were supposed to be managed. An unnecessary referral was assumed to be a woman without obstetric complications who should have received care at lower levels of care including health centers. Additional data included transport costs, clinical care costs and accessibility to LCRH in terms of distance and modes of referral. These were retrieved from obstetric records, ambulance call logbooks, referral registers, perinatal registers and nancial records.
Mode of referral was grouped into ambulance transfer and self-referral. Ambulance transfer was a woman brought by ambulance and escorted by health care professionals to LCRH. Self-referral was a woman who arrived in LCRH by other modes of transport such as private car, motorcycle or by foot directly from home. Associated conditions are diseases or conditions that may be relevant to a severe maternal outcome but are not part of the chain of events leading to that severe maternal outcome. Our hypothesis was an increased number of ambulance transfers among women managed for either MNM or PLTC as compared to women without obstetric complications at onset of referral who would opt to use other modes of transport.
Costing analysis was performed from the health care provider's perspective, being the county government of Bomet [17]. All costs falling under the county health system were included. These included ambulance and health services provided in LCRH. Our study did not take into consideration opportunity costs when accessing health care. These included costs incurred by patients and families, resources used by health care providers from referring agents and other parties like insurance companies and donors.
Recurrent costs included drivers' and paramedics' salaries, allowances for accompanying drivers and paramedics on night/weekend calls, length of hospital stay, laboratory tests, radiology procedures, clinical care costs incurred during vaginal birth, surgical interventions such as cesarean section or laparotomy and management of other obstetric complications. Recurrent costs also included ambulance fuel/insurance/maintenance, electricity, water bills, and cell-phone communication. We computed transport costs incurred among self-referrals using Geo-measure area calculator to estimate distance from home or private clinic to reach LCRH, multiplied by costs of fuel per kilometer. We also assumed that women had been accompanied by one person implying transport doubled the costs incurred per woman. Fuel costs for non-obstetric cases were excluded. Costs were calculated based on the National Health Insurance Fund delivery costs [14]. Data on quantities and costs of recurrent and capital overhead goods were obtained as recommended [17]. Overheads were calculated using the allocated shared costs based on time and units of consumption of each shared input [18]. Capital items were physically enumerated and the actual amount of recurrent items was obtained by reviewing general store records, pharmacy, purchase records, hospital's supplies, accounts and Kenya medical supplies agency records. All costs were presented in Kenya Shillings (KES) and converted into US dollars (1 US$ = 103 KES). The cost of capital resources such as ambulances appeared as a single large amount at the beginning of an evaluation period. We computed the equivalent annual costs based on WHO regional recommendations through annualizing capital costs [17]. The costs in each cost center were added to obtain total costs. These were divided by the interventions output to provide the unit cost of delivering costs for necessary ambulance transfers and self-referrals. The unit cost per inpatient per day was obtained by dividing total inpatient costs (capital and recurrent costs) by the total number of admission days for women with necessary referral.
Ambulance or self-referral bene ts were presented as cost per life years gained (LYG) for every referral of women categorized as necessary referrals based on local life-expectancy tables. This was adopted from the WHO reported 66 years as female life expectancy in Kenya. To calculate life expectancy related to different ages of women, we used the table for "the expectation of life at age" [19]. The table does report the average life expectancy for each age; clustered in age groups each containing four years, from 1 to 4, 5 to 9, 10 to 14, until 100+. The older the woman becomes the lower her life expectancy, the lower the life years gained when ambulance referral or self-referral were 'necessary referrals'. The formula to calculate LYG was as follows: LYG for every MNM or PLTC case categorized as necessary referral was 66 (average life expectancy) minus the expectation of life at the age of the mother (as indicated in the expectation of life for age table). For every categorized case as necessary referral, the same formula was adopted.
Therefore, total LYG was the sum of all life years gained for necessary referral cases. Finally, our results were derived from the formulas below: Sensitivity analysis was performed using discount rates of 3% and 6%. First, we discounted costs and life years gained (LYG) by 3% for the proportion necessary referrals and later increased the discount of life years gained to 6%. Finally, incremental cost effectiveness ratios (ICER) for the referral intervention were considered acceptable, attractive, or very attractive interventions when costs per gained life years were below Kenya's GDP per person of 1507.8 USD; attractive when < 150 USD and very attractive when < 30 USD [19].
∝ and ∂ represents necessary and unnecessary referrals Formula used to discount = (1-e (-r*LYsXi))/r) where e is function of exponential, r responds to 3% or 6% discount rate and LYsXi are the life years gained for each mother.

Discussion
Our main ndings reveal that implementing an effective ambulance service for emergency obstetric care in Bomet County was highly cost-effective with $14.8 per life year gained, far below the $30 WHO costeffectiveness threshold [19,20]. Further sensitivity analyses conducted on the rate of necessary referrals, ambulance costs and discount rate emphasized the robustness of this conclusion, similar to other studies in Uganda and Ethiopia, and demonstrating the signi cance of effective ambulance transfer systems for emergency obstetric care in remote African settings [7,[21][22][23].
In our setting, ambulances were mainly used to transfer pregnant women from low to higher level health facilities [21][22][23][24]. Our ambulance obstetric transfer rate of 3.7% was lower than those in other African studies that reported rates between 5% and 66% [21,24,25]. This could be explained by accompanying paramedics on ambulances providing support and triage of patients at the referring health facilities, preventing unnecessary referrals. Nonetheless, still one-quarter of women transferred by ambulance, together with nine out of ten who had come to LCRH as self-referrals, were unnecessary referrals. These women could have been managed in sub-county hospitals and health centers without any use of ambulance services. This highlights the importance of implementing an ambulance obstetric referral protocol, educational interventions in health centers (i.e drills, simulations, continuing professional development), strengthening supervision in lower level health facilities, improving drug availability and equipping lower health facilities with drugs and essential supplies in order to prevent unnecessary interfacility referrals which lead to increased workload in higher level facilities and avoidable costs for families and health systems [26,27].
Main indications for ambulance transfer were prolonged labor and hypertensive disorders, comparable to other studies from LICs [24,28]. This may be explained by poor quality of ANC leading to suboptimal knowledge of obstetric danger signs and low levels of birth preparedness and complication readiness (BP/CR) [29,30]. Health promotion regarding BP/CR at all stages of women's reproductive life with support from community members is needed. The other indication for referral was sepsis/peritonitis following cesarean section. This could be explained by poor sterilization procedures and inappropriate use of antibiotics prescribed by inexperienced medical o cers working in sub-county hospitals, which calls for strengthening of quality and continuum of obstetric care in those health facilities [31].
Our ndings reveal that postpartum hemorrhage was the commonest PLTC, with one case of uterine rupture among the women who had been admitted as self-referrals. This highlights challenges rural women face due to delays in seeking or reaching basic delivery care in low-income countries [7,21]. Most women who had been transferred in critical conditions by ambulance, their referring health centers had called prior for ambulance services. This emphasizes the importance to implement the national standard referral guidelines [32]. There was no difference in terms of maternal morbidities between ambulance transfers and self-referrals which could be explained by: i) reduced transport delays, ii) improved clinical conditions upon arrival due to care by paramedics while on board, or iii) improved obstetric care after arrival in LCRH. Additionally, women who presented as self-referral came from less remote areas than those transferred by ambulance or more often from private clinics. Ambulance transfer thus improved survival of women coming from more remote areas who otherwise might have arrived in poorer condition.
Health centers notably used mobile phones to call for ambulance support, which is much cheaper than VHF radio communication reported in other studies [7,22,23]. Ambulance transfers were free of charge to parturient women delivered to LCRH. The costs were reimbursed by the county government of Bomet through a xed lease agreement with Kenya's red cross society, not affected by increased number of referrals covered per day. There is a need to increase awareness on ambulance services through availing toll free numbers so that local people can call them directly as a more friendly and effective means of referral as compared to the current practice which increases uncomplicated births to LCRH [33,34]. In LICs, mobile phone technology has been associated with signi cant reductions of maternal and neonatal deaths [33]. Thus, there is a need to advocate for nation-wide mobile phone coverage [34]. Other authors have suggested options like motorcycle ambulances based at health centers, as a cheaper alternative to car ambulances based at the district hospital [5]. Unfortunately, this innovative transport approach may not be replicated in every setting due to cultural acceptance issues [5].
This was the rst cost-effective analysis of a referral system comparing ambulance transfers and selfreferrals among women admitted with obstetric complications in a rural hospital. We evaluated effectiveness using the adapted sub-Saharan African MNM criteria as a clinical judgement tool, unlike prior studies which used theoretical judgment, raising concerns about WHO MNM criteria being too restrictive [7,16,22,23]. All ambulances stationed in LCRH had accompanying paramedics who provided support and guidance to women while on board. In terms of limitations, the retrospective nature of this study relied on routinely collected obstetric data, and bias may have occurred because of missing information in the obstetric charts. Our data covered women from remote areas, which render our results and recommendations di cult to generalize to other urban areas in Kenya. Notably, we applied the national life expectancy tables without adjusting for pathologies and regions which may have led to overor under-estimation of bene ts. Unfortunately, speci c life expectancy tables for speci c regions are not available. Ambulance services themselves are not the only factor in uencing referrals for skilled birth attendance. Other contributing factors are good mobile phone network coverage, cheap communication means, free of charge ambulance services, 24-h availability of ambulances and ability for LCRH to provide CEmONC. We focused mainly on survival and did not consider quality of life and disability that may also be relevant. Delayed cesarean section may indeed also impact on quality of life with vesicovaginal stulas and child disabilities as potential complications of that delay [35]; prevention of maternal death may obviously have profound bene ts for the women involved and their families [36]. Although ambulance services were meant for women with obstetrical complications, it is inevitable (and desirable!) that these were also used for other, non-obstetric referrals. For example, lower-level health facilities frequently referred critically ill neonates or children or adults with severe anemia who required immediate blood transfusion. These possible further bene ts were excluded from our analysis but are presumably consistent.

Conclusions
Cost-effectiveness of ambulance services in our setting was very attractive. These can even be made more cost-effective when efforts are geared towards training educational interventions (i.e drills, simulations, continuing professional development) for health care providers, implementing an ambulance referral protocol, improving drug availability and equipping lower health facilities with drugs and essential supplies in order to prevent unnecessary inter-facility referrals.

Declarations
Availability of data and materials All the necessary data and materials are within this manuscript. In case any more data or materials are needed, they are readily are available on request from the corresponding author.

Competing interests
Two of the authors (TvdA and JvR) are members of the editorial board of this journal: TvdA an associate editor and JvR is a section editor.

Consent for publication
Consent to publish this manuscript from the participants was deemed not applicable since the manuscript does not contain any individual person data.
Ethics approval and consent to participate The School of Economics, University of Nairobi rst approved the research proposal and ethical clearance granted from Institutional Research and Ethics committee of Moi University and Moi Teaching and Referral hospital (Approval number: 0003422). While, LCRH administration gave us permission to proceed with the data collection. Written consent was deemed not necessary, since this study was retrospective analysis of routine clinical practice data. The latter ethical committee therefore waived written consent.

Funding
No funding received for this study.
Author contributions RK and MKM conceived and designed study protocol. RK supervised data collection, analyzed and draft rst manuscript for revision by TvdA, LB, JvR and MKM. TvdA, LB, JvR and MKM contributed signi cantly to study design, data analysis, interpretation and manuscript writing. All authors read and approved the nal manuscript.