Patient costs of diabetes mellitus care in public health care facilities in Kenya

Summary Objective To estimate the direct and indirect costs of diabetes mellitus care at five public health facilities in Kenya. Methods We conducted a cross‐sectional study in two counties where diabetes patients aged 18 years and above were interviewed. Data on care‐seeking costs were obtained from 163 patients seeking diabetes care at five public facilities using the cost‐of‐illness approach. Medicines and user charges were classified as direct health care costs while expenses on transport, food, and accommodation were classified as direct non–health care costs. Productivity losses due to diabetes were classified as indirect costs. We computed annual direct and indirect costs borne by these patients. Results More than half (57.7%) of sampled patients had hypertension comorbidity. Overall, the mean annual direct patient cost was KES 53 907 (95% CI, 43 625.4‐64 188.6) (US$ 528.5 [95% CI, 427.7‐629.3]). Medicines accounted for 52.4%, transport 22.6%, user charges 17.5%, and food 7.5% of total direct costs. Overall mean annual indirect cost was KES 23 174 (95% CI, 20 910‐25 438.8) (US$ 227.2 [95% CI, 205‐249.4]). Patients reporting hypertension comorbidity incurred higher costs compared with diabetes‐only patients. The incidence of catastrophic costs was 63.1% (95% CI, 55.7‐70.7) and increased to 75.4% (95% CI, 68.3‐82.1) when transport costs were included. Conclusion There are substantial direct and indirect costs borne by diabetic patients in seeking care from public facilities in Kenya. High incidence of catastrophic costs suggests diabetes services are unaffordable to majority of diabetic patients and illustrate the urgent need to improve financial risk protection to ensure access to care.

diagnosis, poor quality of care or the lack thereof, presence and severity of complications, and comorbid conditions are the most important factors related to DM care costs. 20,21 Evidence of patient costs associated with DM care is needed to assess the economic impact of DM to households, the extent to which DM patients and households are protected from financial hardship due to health care use and to design effective financial risk protection mechanisms for this group of patients. 6,21,22 We therefore conducted this study to document the patient costs of DM at primary care level in Kenya.

| Study setting
The study was conducted from June to December 2017 in two sites in Kenya (Kilifi and Bungoma County) purposively selected to reflect a diverse set of demographic, socio-economic, and geographical settings. Kilifi is located on the coast of Kenya, and a high burden of stroke and heart failure has been described in this area. 23 The population in Kilifi has been well characterized by data from the health and demographic surveillance system run by the KEMRI Wellcome Trust Research Programme. 24 The Webuye Health and Demographic Surveillance System run by Moi University is located in Bungoma County in the western region of Kenya. 25 Multiple cardiovascular risk factors have been identified in this area. 26 Table 1 outlines the study site characteristics.
Six public health care facilities were purposively selected in consultation with county health officials in respective counties to generate a sample of facilities with different workloads, plus the location of the clinics relative to the communities served. However, due to the 150-day nation-wide nurses' strike at the time of data collection, 27 data were collected from five facilities unlike the anticipated six facilities in the two counties. In Kilifi, a public hospital and a health center that provided DM treatment were selected, while in Bungoma, three public hospitals were sampled.
For this descriptive analysis, data from all the facilities were pooled.

| Sample size and sampling
The target enrolment was 282 patients for a sample size sufficient to obtain an estimate of DM patient costs based on the formulae by Kirkwood 28 : where Z α/2 is the critical value of the normal distribution at α/2 (for a confidence level of 95%, α is .05 and the critical value is 1.96), Z β is the critical value of the normal distribution at β (for a power of 80%, β is 0.2 and the critical value is 0.84), P is the expected true proportion of DM in the population in Kenya of 10% (0. 10), and e is the desired standard size of standard error around the estimated proportion of 5% (±0.05). (all data except where otherwise indicated 30 ).
Every DM patient receiving treatment and available at participating facilities during data collection was approached to participate in the study. Patients were eligible if they self-reported DM diagnosis, had received treatment for a minimum of 6 months after diagnosis and were more than 18 years of age. Consenting patients were selected based on meeting the eligibility criteria and the order of arrival at the clinic. Respondents were asked to report on their health service use, associated costs, income, and coping mechanisms if they undertook any of the following to meet DM care costs: borrowing (having taken a loan), selling household items or assets (eg, livestock), and use of savings.

| Measuring patient costs
The cost-of-illness approach was used to document patient costs. 29 Interviews were conducted using a structured questionnaire. Three trained interviewers collected the data in the two study sites following a pilot from a facility in Kilifi that was not selected in the main study. Interviews were conducted primarily in Kiswahili, with local languages (Kigiriama and Kibukusu in Kilifi and Bungoma, respectively) used to clarify questions where necessary. Respondents were asked about costs incurred for different care-seeking episodes described in Table 2.
To annuitize sick visit costs, we summed up costs incurred during current care visit and any reported outpatient visit costs that occurred due to DM in the last 4 weeks prior to the study then multiplied by 13 (assuming there are 52 weeks in a year). On the other hand, to annuitize costs in other care-seeking episodes described in Table 2 Indirect costs were estimated based on the total hours lost while seeking care as well as the cost of illness-related to lost productivity for both patients and their caregivers, assuming that these hours would have been used for  30 Income lost due to DM illness was therefore estimated by multiplying the estimated number of lost production hours due to DM by the official minimum wage; by the official minimum wage of Kenya shillings (KES) 8568 (US$ 84/month) in the agricultural sector in 2017 (given the main economic activities in our study sites). 31 We assumed an average workday of 8 h/day and 22 working days/month. Caregivers' lost productivity was also estimated by multiplying the total number of hours spent caring for the patient by the official minimum wage rate.
Income was estimated by asking detailed questions about income categories, including patient income, income for household members, welfare payments, and government assistance. As a measure of financial risk protection, we compared total direct costs incurred in the overall sample and by socio-economic status, against annual household income and total direct costs excluding transportation costs and defined costs as catastrophic if they exceeded 10% of annual household income. 32

| Data management and analysis
Data were double entered to enhance data quality.   (Tables 4-6).

| Inpatient costs
Twenty-two percent (n = 37) of the patients in the study reported an inpatient admission in the past year with each admission lasting a mean of 1.6 days (95% CI, 1.

| Medicine collection costs
The median number of routine medicines prescribed to diabetes-only patients was 2 (IQR 1-2) and 3 (IQR, 3-4) for comorbid patients.  (Table 7). In addition, more than half (57.7%) of sampled patients reported obtaining their routine medicines from a public hospital ( Figure 1). Medicines accounted for 87.3% of total OOP costs during medicine collection visits while transport and food accounted for 10.4% and 2.3% in the overall sample, respectively (Table 4).

| Diagnostic/laboratory test costs
The main routine diagnostic tests reported by diabetes-only patients were fasting blood sugar (FBS) and weight-          This was closely followed by transport costs in the overall sample and in the comorbid patients while user charges were the second highest cost category after medicines in the diabetes-only group (Tables 4-6). Among patients experiencing catastrophe, patients in the lowest wealth quintile incurred higher direct costs with few resources to meet the health care costs ( Figure 2). Nonetheless, Figure 3 shows a decreasing proportion of patients experiencing catastrophic costs if the 10% annual income threshold was increased. Patients had to borrow (23.3%) from friends/family, sell an asset (29.9%), and use savings (36.8%) to pay for DM care costs.

| Productivity and social impact of diabetes mellitus
Patients were asked to report any work days missed due to DM illness in the last 3 months before the survey. Of the 163 patients, 33% reported to miss a median of 21 (IQR, 7-60) working days. Forty-five percent of the respondents FIGURE 2 Relationship between catastrophic costs and socio-economic status

| DISCUSSION
This study has estimated patient costs for adults with DM that sought care at five public health care facilities in two counties in Kenya. The study's main finding is that patient cost for DM are driven by medicine expenses. Similar findings have been reported by recent studies in Kenya and South Africa. 19,34 Costs due to medicine have been shown to reduce adherence to medication and demand for health services by patients with NCDs. 35 Past studies conducted in LMICs have however shown that social health insurance schemes do not comprehensively cover the costs for medicines 36 and that OOP costs, which are majorly contributed by medicines, are a hindrance to attainment of universal health coverage in many low resource settings. 37,38 Indeed, any reductions or removal of medicine costs is likely to increase access to DM health care services, but additional resources will be required to cover any concomitant increase in service utilization.
The incidence of catastrophic costs documented in this study is arguably high and suggests that DM care in the sampled health facilities is unaffordable to majority of patients, especially those in the lowest wealth quintile whose capacity to pay is limited compared with those in higher socio-economic group. This is a concern given the high poverty rates in Kenya (36.1%) and that only 19% of Kenya's population have a form of health insurance. 39,40 Furthermore, a past study has shown that families with a member with an NCD incurs three times higher costs compared with families without a member with an NCD. 17 Our study has also shown that DM patients reporting hypertension comorbidity incur higher costs overall compared with diabetes-only patients. This places additional financial burden on families of these patients, similar to findings of previous studies. 18,41 Our results indicate that transport cost offers an access barrier to DM patients given that it takes a significant proportion of total direct costs in all care-seeking episodes. In part, the high transport costs reported in this study can be attributed to poor quality of care in public health care facilities. For instance, 48.9% of the patients reported lack of medicines and diagnostic facilities as a reason for not visiting nearest facilities. This phenomenon has been observed from studies in Uganda and Zambia. 42,43 Prior studies conducted in Kenya highlight that transport costs are a key access barrier especially to poor patients. 38,44 Moreover, 36% of patients in the overall sample reported a sick visit outside of scheduled clinic appointments incurring an annual mean cost of KES 38 597 (US$ 378.4). Failure in health-care delivery has been shown to increase the risk of catastrophe, exacerbate socio-economic iniquities, and reduce the probability of comprehensive treatment. 45 These findings therefore reiterate the need for policymakers to develop mechanisms to improve quality of care for diabetic patients in public health facilities since this has serious cost implications on patients. Additionally, since more than three quarters (76.7%) of the patients reported attending their routine scheduled clinics monthly, introducing mechanisms to minimize facility visits, for example, by enhancing and supporting self-care by patients is likely to reduce transport costs.
The indirect costs due DM care in this study are noteworthy. For instance, of the 163 patients enrolled in the study, 48 (29.5%) had to stop working because of DM and 54 (33%) reported a median loss of 21 working days over the past 90-day period, which is equal to 1 month's wage in the informal sector in Kenya. Similarly, the overall mean indirect costs in all care-seeking episodes was KES 23 174 (US$ 227.2) and were primarily contributed by long waiting times at health facilities. It has been shown that long-waiting times while receiving care reduces demand for chronic care services. 46 To achieve optimal efficiency and increase demand for service delivery for patients with chronic illnesses like diabetes, there is urgent need to redesign health service delivery for these patients with a view to making care more patient-centred to meet the unique needs of these patients.

| LIMITATIONS
Our study had some limitations. Due to the recall periods, results may be subject to recall bias. Our study only focused on those who utilized health services, thus excluding those with undiagnosed diabetes who may still incur costs due to symptoms associated with diabetes. Also, costs reported in our study could be potentially overestimated since patients were selected purposefully. Consequently, our findings are not nationally representative. Also, our study relied on the diagnosis reported by patients, hence could not distinguish between type 1 and type 2 diabetes. Use of an official minimum wage to estimate productivity losses for all patients could have potentially overestimated indirect costs among patients who were unemployed prior to their illness or underestimated indirect costs among those who were employed. 30 These limitations notwithstanding, the findings presented are potentially useful as inputs in costing and/or cost-effectiveness models that require patient cost and suggest there are significant OOP costs associated with DM management in public facilities in Kenya, which offer a barrier of access to care.