Heterogeneity in Average Cost per Patient with type 2 Diabetes at Primary Health Facilities in Mexico: Comparing Comprehensive Diabetes Management Medical Offices with General Practice


 Background. Despite the high health and financial burden imposed by diabetes in Mexico, few studies have estimated the cost per patient treated. The objective of this study was to estimate the average annual cost per patient (unit cost) with diabetes among 60 primary health facilities in Mexico comparing comprehensive diabetes management medical offices (MIDE) and those from general practice (Non-MIDE). Methods. We described the variation in unit costs across these two types of medical offices and explored factors associated. Unit costs were the sum of staff, medications, laboratory tests, and equipment. We show descriptive statistics to analyze the heterogeneity of unit costs, and the distribution of total costs by input and the distribution of staff costs by personnel all by medical office. We estimated a multivariate linear regression model to explore factors associated with the unit costs. Results. Unit costs vary from $267.2 USD in Non-MIDE offices to $410.6 for MIDE. Unit costs were negatively associated with scale, Non-MIDE offices, medical competence, patient knowledge of diabetes and positively associated with comorbidities. Conclusions. Results from this study might help design more efficient programs for diabetes care in primary health facilities to reduce the burden of diabetes in the system. Investing in staff training and educational interventions to increase patient knowledge of diabetes could be promising interventions to reduce diabetes care costs in primary care settings.


Background
Type 2 diabetes is a public health problem worldwide and one of the leading causes of death (Barraza-Lloréns M 2015). 80% of diabetes deaths occur in low and middle-income countries where the prevalence has increased more rapidly in the past 15 years (Abegunde DO 2007). In Mexico, the burden of morbidity and mortality associated with diabetes is high, ranking seventh among the ten countries with the highest number of people with diabetes in the world (8.7 million in 2013) (Abegunde DO 2007;Wild S 2004). It is a growing financial burden for the health system, mainly due to the treatment of complications (Stevens G 2008).  (Barquera S. 2013). Finally, no studies have analyzed differential costs of alternative delivery models such as comprehensive diabetes management or multidisciplinary risk assessment and management programs for patients with diabetes. A better understanding of heterogeneity in costs across health facilities and its determinants is essential to identify potential savings and strategies to achieve metabolic control at lower costs.
The second largest segment of the Mexican Social Security system is the Institute of year) and improving self-care (Díaz I. 2011). Although MIDE treats any patient with diabetes, those with a clinical condition that reflects an inadequate response to previous treatments receive priority (glycosylated hemoglobin equal or greater than 7% (53 mmol/mol) or pre-prandial glycemia greater or equal to 130mg/dl). MIDE medical offices are located-along with general practice offices-in primary health care facilities.
The objectives of this study were: 1) to estimate the average annual cost per patient with type 2 diabetes (unit cost) per facility among 60 ISSSTE primary health facilities comparing MIDE and general practice (Non-MIDE) offices, 2) to describe the cost per patient variation across medical offices, and 3) to explore factors associated with the average annual cost per patient with type 2 diabetes.

Methods
The study was approved by the Institutional Review Board (Ethics Committee) at the National Institute of Public Health in Mexico (IPF Code 3627801) and the Ethics Committee at ISSSTE.

Study design
There are four types of primary health facilities in ISSSTE: hospital clinics, family medicine units, family medicine health facilities (FMHF) and specialized family health facilities (SFHF). We selected FMHF and SFHF because they provide care to 62% of beneficiaries and 61% of MIDE medical offices at the time of data collection. SFHF are different then FMHF because some of them have medical specialties such as general surgery, gynecology and obstetrics, internal medicine and pediatrics, among others.
We selected 60 health facilities for the study using a two-step stratified sampling strategy.
The two strata were FMHF and SFHF. We selected 50 FMHF and 10 SFHF; including the six largest FMHF that have at least 150,000 beneficiaries each one and four SFHF units with more than 100,000 beneficiaries, with probability 1. We selected the remaining facilities with probability proportional to the size of their catchment areas in both strata.
Given the distribution of the facilities in the country, Mexico City concentrated 40% of the sampled facilities.
All health facilities in the sample had a general practice office (from now on called Non-MIDE), and 55 had a MIDE office. However, we excluded two Non-MIDE offices and three MIDE offices due to lack of data on the number of outpatient visits or missing values in the vignettes. Thus, the analytical sample consisted of 58 Non-MIDE and 51 MIDE medical offices (Figure 1).

Measurements
Data for this study were collected retrospectively for each month of the entire 2015 fiscal year. In table 1 we gathered information on outputs, inputs, and quality of services using five questionnaires. performed by the physician in the medical appointment and included the same items as the vignettes for process quality; and questions related to the patient's knowledge of diabetes.

Estimation of unit costs
For each facility, we estimated the unit costs as the sum of the total annual costs of staff, medications, equipment, and laboratory tests, divided by the number of outpatient visits or patients: Where UC is the unit cost at facility j, medical office k (MIDE or Non-MIDE); IC represents the total cost of each of the four inputs i at facility j, medical office k, for the following categories: staff, medications, laboratory tests, and equipment. O is the output measured either as the number of diabetes visits or the number of patients with diabetes.
We added the monthly outpatient diabetes visits over the entire year to obtain annual estimates. We divided the yearly number of outpatient visits by the average number of medical visits per patient as reported in medical records to estimate the number of patients with diabetes that received care during 2015.

Staff costs
The annual cost of staff was estimated as follows:

Cost of medications
We estimated the total annual cost of medications using information from medical records on medicines prescribed and prices of each medicine. The medications, which included both oral and injected, were used for the management of type 2 diabetes, including medications for hypertension and dyslipidemias experienced by patients with diabetes.
Clinical files recorded doses or changes in medication or doses only the first time the medication was prescribed over the year. We thus assumed that patients used the prescribed medications throughout the year, i.e, we imputed all prescriptions and changes between visits. Medication costs were estimated as follows: Where DC is the annual cost of medicines for patient i, in facility j, and medical office k. T represents tablets or insulin doses and B is the number of tablets or insulin doses per bottle.
P is the price of each bottle for medication m.

Equipment costs
The survey included information on the availability of medical equipment. To estimate the annual cost of medical equipment we used the following formula: Where EC is the total annual cost of equipment that belongs to health facility j, medical office k. E is the number of equipment c and P its price. If the equipment was with other offices, we divided the cost based on the proportion of outpatient visits in each office.

Laboratory tests costs
We first estimated the number of laboratory tests per visit, which is the number of laboratory tests per patient divided by the number of visits per patient in the year. We then multiplied the average number of laboratory test per visit by the average outpatient visits in each facility and the price of each test: , , =1 * , + Where LC represents the annual costs of laboratory tests in facility j, and medical office k; L is the number of tests per patient i, v is the number of visits per patient, V is the number of outpatient visits at the facility level, and P is the price of each laboratory test x.

Composition and distribution of unit costs
To analyze the variability and heterogeneity of the cost per visit and cost per patient (unit costs), we show descriptive statistics of these two indicators by facility and medical office.
We also present the distribution of total costs by input: medicines, personnel, laboratory tests, and equipment, to illustrate the relative contribution of each category. Similarly, we show the distribution of staff costs by personnel.

Factors associated with facility-level variation of diabetes care costs
To explore factors associated with facility-level variation of unit costs, we first present the correlation between unit costs and scale (number of patients with diabetes) for MIDE and Non-MIDE offices on a log-log scale, using a locally weighted scatter plot smoothing test for non-linear associations.
We then fitted a multivariate linear regression model to explore factors associated with unit costs at the medical office level. The dependent variable was the facility-level unit cost, log transformed as the costs are not normally distributed. We included in the model several variables categorized under supply-and demand-side characteristics ( Table 2). complications (nephropathy, retinopathy and neuropathy). As MIDE and Non-MIDE offices are located in the same health facility, we used robust standard errors clustered at the facility level.    Figure 2 illustrates the distribution of total costs and staff costs. Medications represent the largest proportion of total costs (higher for MIDE) followed by staff salaries. Within staff costs, general practitioners and medical specialist represent more than 50% of total staff costs.

Fig. 2. Total diabetes costs and staff cost breakdown
We explored the association between unit costs and scale (number of patients) in Figure 3.
First, the graph shows a negative association between costs and scale. Second, in general,

*T test for difference of means
In Table 5 we explored the influence of the characteristics presented in Table 4 on unit cost variation. We present the results of three regression models incrementally adding more variables. Model 3 explained 45% of the total variation in unit costs across facilities. A 10% increase in the number of patients was associated with a 19% reduction in unit costs.
Unit costs in Non-MIDE medical offices were 19% lower compared with MIDE. A higher competence (quality process from vignettes) score was associated with 26% lower costs compared with a lower competence score. Higher patient knowledge of diabetes is associated with lower costs. The proportion of patients with neuropathy was associated with higher costs. We also found that higher patient knowledge of diabetes was associated with lower costs.
This result is consistent with a review of studies that evaluated the effect of educational interventions for patients on costs (Boren S.A. 2009). In those studies, the interventions were associated with higher treatment adherence, better compliance with the medical appointments and frequency of laboratory tests, which in fact resulted in better patient control and therefore in reduction of complications.
Our results also showed that a higher level of competence reduces costs. Despite these limitations, as of our knowledge, this study is the first to estimate unit costs based on primary data and to explore factors associated. Findings from this study might be translated into recommendations to improve efficiency in care. For instance, given our findings that a higher level of competence and a higher patient knowledge of diabetes reduces costs, investing in staff training and patient educational interventions could be promising interventions to reduce diabetes care costs in primary care settings

Conclusions
We identified variation in unit costs of two models of care for diabetes patients. In average unit costs for MIDE were almost twice as expensive as Non-MIDE, due to MIDE offices offer more specialized and personalized services compared to Non-MIDE offices (general practitioner offices). The costs had a statistically significant and negative association with the number of patients, similar to other studies showing economies of scale. We found that higher the level of knowledge of diabetes by patients, there is an association with lower costs. In the same direction a higher level of competence of the health care personnel is