To the best of our knowledge, our study is the first to evaluate the effect of the Sanming model of systemic hospital reform on inpatient expenditures for the rural population using 10 years of real-world insurance claims data. Although in this study, a large sample use of administrative data allows study of real-world utilization patterns in an unselected population without nonresponse or recall bias problems, it should be noteworthy that a large sample easily produces significant results. So, we not only interpreted the results based on significance, but also interpreted it in light of the actual magnitude of the effect size, considering the practical conditions.
The average growth rate of total inpatient expenditures after the reform decreased by 1.34% (5.38% vs 6.72%), compared with the period before the reform. A previous study  identified four primary targets for cost containment: price controls, volume controls, budgeting and market oriented policies. Looking at the specific polices of the systemic reform in the Sanming model, we can find this reform refers to all these four targets, such as price controls through the “Two Invoices” system, volume controls through improving appropriateness of prescription medicines to prevent overtreatment, budgeting through health sectoral budgeting, and market oriented policies including payment reform of FFS model and DRG. The combination of all these cost containment targets in the systemic reform of the Sanming controlled its soaring hospital expenditure, which provide critical lessons for China and other countries with similar issues.
The amount of OOP expenditures after the reform showed a downward trend. Holding all else constant, individuals after the reform spent ¥308.42 less on OOP expenditures (p < 0.001) than before the reform. It was also shown that OOP expenditures had a similar significant decrease regardless as to whether the fixed effects of the year and hospitals or cluster standard errors by hospital were included in the model.
Without cluster standard errors according to hospital to account for intrahospital correlation, length of stay had a significant decrease after the reform (0.73 days, p < 0.001); however, the decrease of length of stay was no longer significant after cluster standard errors by hospital were included (p > 0.05). These results revealed that a slightly shorter length of stay was not mainly contributing to the savings of OOP expenditures. The OOP expenditures slowdown was largely accomplished through the systemic reform strategies.
The absolute amount of OOP expenditures decreased, however, OOP expenditures as a share of total expenditures after the reform was still 41.05–47.60%, higher than that in high-income countries which is generally below 30% . It is important to study which are particularly likely to have more OOP expenditures to provide stronger risk protection for the particular population.
The results in the GLM analysis showed that OOP expenditures was higher for older people. Previous studies also indicated that older people experienced higher hospital expenditures [25, 26]. The high expenditures we found for older people and the resulting increased financial burden are particularly relevant today given China’s aging demographic. In 2016, people aged over 60 years accounted for 16.7% of the total population in China . Our study showed that although the government had increased the health investment for rural populations in recent years, those vulnerable people, such as the elderly, were still not better protected. Hence, the health financing structure should concentrate on equity of China health financing.
The results in the GLM analysis also showed that individuals, hospitalized at high level of hospital and in other metropolitan areas, had significantly higher OOP expenditures. These two items share a common thread of reflecting more severe diseases, which usually need hospitalization at high level of hospital and in other metropolitan areas, resulting in more total inpatient spending and more OOP expenditures.
In our study, it is noteworthy that the official reimbursement rate and the ceiling of annual compensation had the opposite effect on OOP expenditures. The OOP expenditures rose as the official reimbursement rate increased, indicating that raising the official reimbursement rate alone couldn’t reduce the financial burden for the rural population. In China, the actual reimbursement rate was lower than the official reimbursement rate, because many services provided in the hospitals are not covered by NCMS and patients have to pay 100% for those uncovered services. Take 2016 as an example, the official reimbursement rate had been above 85%, however, the actual OOP expenditures as a share of total expenditures was 41.05%, implying that many medical services were still uncovered by NCMS. Compared with raising the official reimbursement rate that only covered a part of services, increasing ceiling of annual compensation per patient can better protect financial risks for the rural population. Other studies [9, 27] also showed that the insurance scheme with co-payment couldn’t protect poor people from becoming medically impoverished. This result is important for policy-making. For the public health insurance scheme, decision-makers should focus on how to provide patients with better financial risk protection and prevent medical impoverishment, while ensuring the sustainable development of the health insurance fund.
There are several limitations to this study. Firstly, although we used 10 years of data for our analysis, these findings cannot be interpreted as the long-term effect of the reform. Secondly, since we couldn’t obtain the data about the medical quality of primary hospitals, although the medical quality of secondary and tertiary hospitals as the main providers of inpatient medical services wasn’t impacted (this was not reported at the result part, see Additional file 1), it is not for certain whether the systemic reform had reduced the medical quality of primary hospitals. Thirdly, we only examined the effect of hospital reform in Sanming, therefore, the results may not be generalized to other areas. Consequently, policy makers and health care workers should treat the empirically results with caution, when applying the Sanming model in other contexts.