Analytical framework

This study uses a health systems framework [16,17] to guide the analysis ( Box 2 ). The framework builds on earlier approaches used to analyze health system reforms [18,19]. The national Healthcare Development Strategy 2008–2017 follows a similar structure to the health systems framework used in the study, enabling the analysis of the changes in health system goal of financial protection following health system reforms aimed at expanding insurance coverage, and exploring the association of insurance coverage with service utilization, OOP payments incurred and forgone health care.

Data sources

Two main sources of data were used for the years 2006–2012. First, publically available datasets from the Moldovan National Center for Health Management (CNMS) were collected. CNMS collates data from public health care provider reports. We extracted health service utilization information relating to number of hospitalisations, average number of visits per person, and emergency calls per 1000 residents.

Second, the monthly Household Budget Survey (HBS) was used. The HBS is based on an internationally validated survey and is undertaken by the Moldovan National Bureau of Statistics (NBS) [7]. The HBS is nationally representative and is undertaken through two–stage sampling based on regional areas and a random selection of households. Approximately 5500 households (15 000 individuals) are surveyed annually on a wide range of questions relating to the economic situation of the household and individuals. Responses for the years 2006–2012 were obtained from the NBS. We selected questions relevant to this analysis including demographic, socio–economic, health and health care–related questions. Our outcome variables of interest were calculated from survey questions: “If individuals currently have health insurance”; “if individuals paid for any service (inpatient or outpatient) either formally or informally when using care in the last four weeks”, and “if you were unwell in the last four weeks, but did not use healthcare”. Because of issues of non–response (up to 40% in some years), the age and gender distribution of the HBS was compared to national population data (from NBS [7]) showing high similarity.

Analysis

Using the health system analytical framework [14,15], we examine elements of financial protection in the context of the health system objectives. Equity is a key health system objective for this analysis. We examine equity in insurance status, OOP payments, and foregone health across demographic groups. We demonstrate the interactions between insurance status and equity in other financial protection elements such as OOP. Additionally, we also explore the health system objectives of responsiveness in terms of preferred health care provider, effectiveness through forgone health care, and efficiency in terms of national utilization trends. We triangulate these findings to understand how factors contributing to financial protection are being met.

Descriptive analysis

CNMS data on health care utilization are shown over time. Individual responses on preferred of health care provider, for those with health care use in the last four weeks, were stratified by provider and year.

For insurance coverage, CNMS data was compared with individual HBS responses (stratified by employed status and age groups) over time.

Insurance coverage trends were compared between individual HBS responses and from administrative data (CNMS) sources. Furthermore, we stratified the insurance status of the respondents in the HBS by occupation and age group.

The percentages of individuals reporting OOP payments were described by consultative services, inpatient services and drugs, and by each year. Mean incurred costs by individuals were shown. Additionally, mean costs were compared to average monthly earnings.

Logistic regression

We employed logistic regression to calculate the likelihood of: being uninsured, incurring an OOP for medicines or any health care service from health care used in the last four weeks, and not using (foregoing) health care when reporting a health problem in the last four weeks. Logistic regression was employed as the most appropriate method for binary outcomes.

Covariates from the HBS survey were used to control for and highlight explanatory factors. We included in all models: age group (0–17 years, 18–24, 25–34, 35–49, 50–59, 60–74 and 75+); gender; chronic disease status (yes or no); employment status (employed, self–employed (non–agriculture), self–employed (agriculture), unpaid family worker, unemployed (including those not of working age); and educational attainment (pre–school or no education, primary, secondary, and college or university). We also included year (2006–2012) to look for time trends.

For the regressions on the likelihood of an OOP and forgone health care, we included disability (yes or no) and uninsured (yes or no) as covariates. Additionally, for the regressions on the likelihood of an OOP we included first choice of health care provider (family doctor’s office, home visit, polyclinic (health center), hospital or other (eg, pharmacy). For the regression of foregone health care, we only examined the years 2009–2012, as the question was not in earlier surveys. Analyses were carried out at the individual level, with adjustments for the clustered nature of the survey. All individual responses were included for analyses, except the likelihood of being uninsured. The regression was restricted to those aged over 18 and under the age of 60 years, as individuals outside these ages are eligible for free insurance coverage. We report adjusted odds ratios and 95% confidence intervals. Interaction terms between covariates and year (linear trends) were additionally tested.

Service utilization

We examined average health care utilization rate at the national level and preferred first contact provider from HBS respondents. The health care utilization rate at the national level rose between 2006 and 2012 ( Table 2 ), with the hospitalization rate (per 100 residents) increasing from 16.7 in 2006 to 18.4 in 2012 and the average number of outpatient visits per person rising from 6.02 to 6.45. These numbers are broadly in line with European averages (of 18.04 hospitalisations and 6.93 outpatient visits per person in 2012) [5].

Responses in the HBS from individuals who sought health care in the last four weeks show the majority sought health care in former polyclinics (centers that in majority of cases include both family doctors and out–patient specialists) (48.0% in 2012) with family doctor offices (32.6%), home visits (8.9%) and hospitals (5.2%). These trends have remained fairly constant since 2003 ( Figure 1 ).

Figure 1.  Choice of health care provider for individuals who sought medical care in the last four weeks (2006-12). Shows responses from individuals who sought healthcare in the last four weeks and their first preferred healthcare provider. Those seeking care at a hospital are likely underestimated due to the infrequency of events and small samples employed. Polyclinics include family doctors and out–patient specialists, especially in larger cities and urban centres.

Insurance coverage

Approximately 75% of HBS respondents reported having health insurance during the period 2006–2012 ( Table 3 ). National statistics suggest rising coverage of health insurance from 76.1% to 80.3%. In 2012, only 60% of the working age population (18–59 years) had insurance coverage, with the self–employed and those working in family business having lower coverage levels: 24% of self–employed agriculture workers, 57.8% of self–employed non–agriculture workers, and 53.8% of unpaid family workers had insurance. In 2012, the main reasons for being uninsured were unemployment (27.9%), cost (26.2%), “they would pay for healthcare anyway” (13.9%), working informally (13.1%), “belief that it would not be needed” (8.5%) and working abroad (8.1%).

The odds of not having health insurance were examined using multivariate logistic regression ( Table 4 ). Age was an important determinant of being uninsured, with those aged 25–43 and 35–49 years of age, respectively, 2.9 (OR = 2.898, P < 0.001) and 2.3 times (OR = 2.261, P < 0.001) more likely to be uninsured than those aged 18–24 or 50–60 years of age.

Females were less likely to be uninsured than men (OR = 0.599, P < 0.001), as were those with chronic health conditions (OR = 0.303, P < 0.001). All categories of employment were substantially more likely to be uninsured than those employed. The self–employed, and unpaid family workers were all more than 20 times more likely to be uninsured. The unemployed were 5 times more likely (OR = 4.915, P < 0.001). All categories of education were more likely to be uninsured than those with only pre–school or no–education with primary, secondary, and college or university educated individuals 7.7, 6.1 and 3.2 times more likely to be uninsured.

Over time, there was no evidence of a trend in change in likelihood of being uninsured. Models with interactions with time (results not shown) suggest that the likelihood of being uninsured declined over time for those with chronic conditions (compared to those without), females (compared to males), and the self–employed (agriculture) and unpaid family workers (compared to employed). There was an increased likelihood in being uninsured for the unemployed over time (compared to other employment categories).

Private and out–of–pocket expenditure

Around 15% of HBS respondents who sought health care reported a payment for consultative services and 2% for inpatient care in 2012 ( Table 5 ) compared with 20% and 3% in 2006 respectively. The average costs for both consultative services and medicines had increased since 2006.

The factors affecting the likelihood of an OOP payment for both medicines and health care services were analyzed separately using multivariate logistic regression ( Table 6 ).

Although low numbers and fluctuations in reported costs make conclusions difficult, there appears to be only modest reductions in costs incurred relative to average monthly earnings over time ( Table 7 ).

Out–of–pocket payment for drugs

Older individuals (aged over 50 years), who sought health care in the last four weeks, were more likely to incur a cost than those aged under 34 years. Females were 13% more likely to pay (OR = 1.133, P = 0.013), while those with chronic conditions were nearly 1.9 times as likely to incurring a cost (OR = 1.911 P < 0.001).

Unpaid family workers and the unemployed were less likely to incur costs for medicines, as were those with primary and secondary education. Those without health insurance were 1.3 times more likely (OR = 1.297, P = 0.004) to incur an OOP for medicines compared to than those with insurance. Individuals were only less likely to incur a cost for medicines at hospitals or other services (such as pharmacies).

Over the period 2006–2012, the odds of incurring a cost for medicines have increased (up to 1.8 times more likely to incur a cost in 2011 (OR = 1.772, P < 0.001) compared to 2006. The only significant interaction with time (results not shown) was insurance status with those without insurance increasingly more likely to pay for medicines OOP over time.

Out–of–pocket payment for health care services

Individuals who sought medical care in the last four weeks and incurred OOP payments (informally or formally) for consultative or inpatient services showed different trends to costs for drugs. Older individuals and those under 18 were least likely to incur a cost than those aged 18–24. Those with chronic conditions were 1.3 times more likely to pay (OR = 1.296, P < 0.001), but those with disability were nearly 40% less likely to pay (OR = 0.634, P < 0.001).

Uninsured individuals were 3.8 times (OR = 3833, P < 0.001) more likely to incur an OOP payment for health services. There was no difference between employment categories while only those with college or university education were more likely to pay (OR = 1.391, P = 0.008).

Those using hospitals and polyclinics respectively were 7.2 times (OR = 7.205, P < 0.001) and 2.8 times (OR = 2.723, P < 0.001) more likely to incur a cost than at family doctor offices. Home visits were less likely to incur a cost than family doctor office (OR = 0.415, P < 0.001).

There was a general decline in the likelihood of incurring an OOP expenditure cost for services, although the difference was not significant every year. The only interaction with time (results not shown) was age with those aged 50–60 years having an increased likelihood of incurring a cost over time compared to other ages.

Foregone health care utilization

Using multivariate logistic regression the likelihood of not visiting a health care provider when being sick in the last four weeks was examined ( Table 8 ). Older individuals were more likely not to seek health care with those 75 years or more 2.4 times (OR = 2.379, P < 0.001) more likely than 0–24 year–olds. Females were 9.8% less likely to forego care (OR = 0.902, P = 0.003).

Employed and self–employed (non–agriculture) individuals were most likely to not seek health care when sick. The disabled were less likely to forego health care when sick (OR = 0.819, P < 0.001). Those without insurance were 1.7 times (68%) more likely not to seek health care when sick (OR = 1.680, P < 0.001).

Over time there was an increasingly likelihood of forgoing health care when sick with individuals on average 1.3 times (30%) more likely in 2012 than 2009. There were no significant interactions between covariates and time suggesting the increasing trend in likelihood of foregoing health care applies to all individuals.

Strengths and limitations

The HBS only provides limited insight into financial protection in Moldova over time. As a survey, certain populations–eg, traveling communities or homeless–will be not represented. The high non–response rate ( ~ 40%) raises the issue of reliability of the survey, although the concordance of results with earlier studies and representativeness of age and sex distributions could mean the impact of the high non–response rate is low. There are also potential issues of recall or selective reporting possible. There are limitations in the questions asked in the survey. For example, we do not know individuals’ utilization patterns of health care–only if they have used a health care provider in the last four weeks. Healthcare utilization could confound the results reported here. None–the–less, while we must acknowledge certain groups (eg, those with chronic conditions, older people or the disabled) are likely to use health care more, true financial protection should be equitable across groups. Further research into the exact costs and fees incurred by individuals and explanatory demographic variables is also needed.

Other data sources employed–including CNMS–are from administrative sources that may be prone to errors and issue of data quality. Additionally, as administrative data are only from public facilities, this study is not able to analyze private providers in the health system.

The analyses undertaken may also introduce potential errors. Logistic regression identifies associations through predictive probabilities between groups and outcomes, but cannot prove causality. Our results must be interpreted with this limitation in mind. Additionally, assumptions about grouping of outcomes–including OOP from informal and formal sources and between different services–may obscure finer trends and patterns, but due to small sample size it was not possible to analyze these subgroups separately. Even so, we use multiple covariates not to just examine potential inequities between demographic groups, but also to control for potential effects and elucidate clearer associations than descriptive trends. This allows us to identify whether it is age or employment status, for example, which is the stronger determinant of an outcome. Furthermore, we take into account the clustered nature of the survey design further strengthening the validity of our findings.