The Impact of a New Case-Based Payment System on Quality of Care: A Difference-in-Differences Analysis in China

Introduction

The prospective payment system (PPS) is widely used worldwide to contain rapidly increasing inpatient costs and avoid unnecessary services, and case-based payment is one of the popular methods.^1,2 Total expenditure per hospitalization in public hospitals in China has doubled from 5856.2 RMB in 2009 to 11468.6 RMB in 2022 (exchange rate: 6.74 RMB=US $1.00 in 2022).^3,4 Therefore, case-based payment has been piloted nationwide over the past three years, replacing the previously dominant fee-for-service (FFS) system. In addition to the traditional Diagnosis-Related-Groups (DRG) payment, China also developed a new case-based payment known as the “Diagnosis-Intervention Packet” (DIP) payment. The concurrent implementation of DRG/DIP payment in China is also referred to as the “dual-track” arrangement.⁵ The DIP payment is distinguished from DRG payment mainly due to its more granular classification. In the DIP payment system, patient classification is based on the data-driven combination of the principal diagnosis and procedures, which could yield thousands of groups, while there are usually no more than 1000 groups in the DRG payment system. DIP groups with the same diagnosis while higher treatment intensity, ie, higher resource consumption, will receive higher reimbursement. With more straightforward classification rules and less involvement of clinical experts compared to DRG payment, DIP payment has rapidly rolled out in China.⁶

The premise of the same payment for a given group in the case-based payment is the similar healthcare quality across different providers. However, the payment system itself also has an important impact on the quality of care.⁷ DIP payment could change the quality of care by influencing provider behaviors in two pathways (see a diagram of the analytical framework of our study in Figure S1).^8,9 First, as a PPS, the payment amount cannot be influenced by the service providers for a given DIP group, which is similar to the incentive under the DRG payment, and it may incentivize providers to select more profitable diseases or patients to minimize the cost. If hospitals admit patients with low-risk diseases or less severe patients with lower costs, their quality of care may be improved. Second, providers may also adapt their treatment strategy. To minimize costs, providers may also reduce services, even necessary ones, in a specific DIP group.¹⁰ On the other hand, under the granular classification system with higher payment for groups with higher treatment intensity, which is a major difference from DRG payment, providers may increase the intensity of care by providing sophisticated treatment to maximize their revenue.^11,12 This hospital response is expected to be easier to implement and thus much stronger than in the DRG payment. However, the impact of such behaviors on the quality of care is unclear, depending on whether the behavior changes are moving towards or away from the best clinical practices.⁸

Previous studies provided less empirical evidence on the association of case-based payment with the quality of care compared to costs and efficiency, especially in low- and middle-income countries.¹³ Furthermore, these studies have inconsistent results.¹⁴ Some studies found a decrease in the in-hospital mortality rate following DRG payment reform,^15,16 while others indicated no significant change.^17–20 Some found the readmission rate increased,^15,16,20 yet others reported no significant change or even a decrease.^{14,17,19,21–24} In addition to different study populations and designs, the differences in the case-based payment policies and the policy context may largely contribute to the variation of study results. Specifically in China, studies generally concluded that DRG payment had no significant impact on the in-hospital mortality or readmission rate.^13,25 For DIP payment, one study found no significant change in the operation-associated infection rate.⁶ Two other studies found no change in readmission volume or in-hospital mortality rate.^26,27 However, they merely incorporated quality measures as a part of the overall analysis and did not systematically examine the association of DIP payment with the quality of care or explore its underlying reasons. Therefore, it is crucial to comprehensively understand the quality changes and how they were achieved through hospital responses after the new DIP payment reform in the context of China. The findings of this study will reveal if unintended consequences regarding health outcomes emerged after the payment reform and indicate the provider behaviors to be closely monitored with the nationwide implementation of the DIP payment to avoid the deterioration of patient benefits. Since the DIP payment has the potential to be easily implemented and rapidly rolled out in resource-limited areas,⁶ the results of our study could also yield implications for the payment system reform in other low- and middle-income countries.

Therefore, the objective of our study is to examine the impact of DIP payment reform on the quality of healthcare in hospitalized patients covered by basic medical insurance in a large pilot city in East China, which is one of the first DIP-pilot cities, and explore potential mechanisms of quality change to provide evidence for future payment system reform in China and other countries. We first observed the overall impacts of DIP payment reform on the quality of care. Then, we conducted a heterogeneous effect analysis among different hospital levels. We then examined the temporal dynamic effects after the reform. We also did additional analyses to explore the possible reasons for quality change.

Materials and Methods

Setting

Our study city is one of the largest and most developed cities in China. In 2022, it has a 19.9 million insured population, with the majority (16.2 million, 81.6%) covered by the employee basic medical insurance scheme (EBMIS).²⁸ Before the case-based payment reform, the city implemented FFS payment with a hospital-level cap.

The city started to pilot DIP payment with a hospital-level cap on July 1, 2019. The first batch of pilot hospitals included two public tertiary hospitals and all secondary acute care hospitals in two districts, with hospitalized EBMIS patients as the target population. Using the discharge data in the past three years, over 14,000 DIP groups were generated based on the combination of the first four digits of the ICD-10 (International Classification of Diseases, Tenth Revision) code of principal diagnosis with the ICD-9-CM-3 (International Classification of Diseases, 9th Revision, Clinical Modification) codes of procedures. Relative weight (RW) was calculated by dividing the average expenditure of a certain group by that of all discharge cases.

Data and Study Population

We included hospitalized EBMIS patients in public tertiary and secondary hospitals as our study population. Some secondary hospitals have been promoted by the local health authority to tertiary, but are still regarded as secondary by the payer. Thus, we further divided secondary hospitals into promoted secondary and other secondary hospitals. The intervention group was EBMIS patients in two tertiary hospitals, two promoted secondary and nine other secondary hospitals, which started the DIP payment pilot in July 2019. The control group was EBMIS patients in eight tertiary hospitals, three promoted secondary and 16 other secondary hospitals, which did not participate in DRG/DIP payment until July 2021.

Using a simple random sampling approach, we considered the representativeness and computational feasibility of the dataset to use a 5% sample selected from the full dataset of deidentified discharge records of EBMIS patients in this city from July 2017 to June 2021. The dataset contains information on patient and hospital characteristics, admission and discharge status, diagnoses and procedures, inpatient expenditure, DIP groups, and corresponding RW at the discharge level. In terms of data processing, we dropped 0.0006% of the discharge records with missing gender information before sampling.

Outcome Measures

We included five primary quality outcomes from aspects of effectiveness and safety.²⁹ For effectiveness indicators, we included in-hospital mortality, mortality of surgical patients, all-cause readmission within 30 days, and readmission with the same principal diagnosis within 30 days. The measure of patient safety was operation-associated complications or adverse events. All measures are binary variables at the discharge level.

In-hospital mortality is commonly used as an important quality indicator in acute care hospitals. It was recognized by discharge status as “death” at the individual level. All-cause readmission is another widely used quality measure. However, only the unexpected readmission following prior care can accurately indicate low quality. To better reflect the actual quality of care, we followed some studies to use an additional indicator of readmission with the same principal diagnosis (defined as the same first three digits of ICD-10 codes),^30,31 as quality problems are more likely to arise in diagnostically related admissions.³¹ Although it may include false positive cases and miss true positive ones,³¹ it could serve as a supplement and mutual verification with the all-cause readmission. The indicator of operation-associated complications or adverse events was aggregated by 19 specific conditions (Table S1) defined by the National Health Commission of China.³²

Statistical Analysis

We used difference-in-differences (DID) approach with linear regression at the discharge level to explore the association of DIP reform with the quality of care. For patient in hospital in time :

(1)

where is the outcome variable; is a dummy equal to one for patients in 13 DIP pilot hospitals (intervention group) and zero for 27 non-DIP pilot hospitals (control group). is a dummy equal to one for discharges after July 1, 2019. The interaction term between and is the variable of interest. is a set of control variables, including sex, age group, principal diagnosis, Charlson Comorbidity Index (CCI, a commonly used measure of patient severity based on secondary diagnoses) group,³³ and interactions between sex and age group or CCI group. is the region and hospital fixed effects, and is the time fixed effects at the year-month level. is the error term, and standard errors are clustered at the hospital-year-month level. We tested the parallel trend assumption by replacing Post in the interaction term in equation (1) with a series of year-month dummies during the study period.

Heterogeneous effects in different hospital levels were performed by subsample analysis in 10 tertiary, five promoted secondary, and 25 other secondary hospitals. We observed temporal dynamic changes by replacing Post in equation (1) with dummies for the first and second years after DIP reform.

We considered other events or policies during the study period and performed several robustness checks. First, we considered the COVID-19 outbreak by excluding data from December 2019 to March 2020, which was the most affected period in this city.^34,35 Second, since the volume-based procurement of coronary stents started in January 2021 and may influence the quality of related conditions, we excluded data on diseases using coronary stents. Third, the study city started to pilot the pay-for-performance in several inpatient services in late April 2021, so we dropped the data in May and June of 2021 to consider its potential influence on the quality of care. Finally, we excluded data from four hospitals as their records of secondary diagnoses were substantially missing, raising concerns about the quality of discharge records – more than 90% of the discharge records had no secondary diagnoses, which was more likely due to the low data quality rather than the low severity of patients.

To explore potential mechanisms of quality change, we conducted additional analyses guided by the aforementioned analytical framework of how DIP payment may change the quality of care by influencing provider behaviors. First, we introduced two additional mortality measures in low-risk groups to provide insights into the disease selection of hospitals and its potential associations with the quality of care. Second, we analyzed the level and trends of CCI and its association with quality measures to investigate the patient selection behavior and corresponding quality change. Third, we analyzed the change in intensity of care (measured by RW) and its association with the healthcare quality. Since the price of medical services has been continuously changing during the study period, which would affect the relative expenditure of each DIP group, we generated an “adjusted RW” by assigning the RW in 2021 to the same group for each year to balance the influence of price change and observe a cleaner change of healthcare intensity, ie, the change of provider behavior. The three pathways were circled respectively in Figure S1. More detailed methods are in Appendix 1 and Table S2.

All statistical analyses were performed using Stata/MP 16.0. Statistical significance was determined by 2-sided P <0.05.

Results

Sample Characteristics

A total of 122,637 discharge records were included, with 43,023 (35.1%) from pilot hospitals. Table 1 shows that the average patient age in DIP pilot hospitals was about three years lower than in non-pilot hospitals, and they both decreased after DIP payment. The proportion of patients with higher severity in pilot hospitals was slightly lower than in non-pilot hospitals, while it experienced a larger increase after the reform. In both groups, more than 50% of patients were in tertiary hospitals.

Table 1 Sample Characteristics of Hospitalized Patients Covered by the Employee Basic Medical Insurance Scheme in the Study City

As shown in Table 1, the in-hospital mortality rate in DIP pilot hospitals was lower and experienced a larger decrease after the reform. The mortality rate of surgical patients, readmission rate, and readmission rate with the same principal diagnosis all decreased in pilot hospitals while increased in non-pilot hospitals. The rate of operation-associated complications or adverse events increased in both groups. Descriptive statistics of components of the patient safety measure and monthly trends of unadjusted outcome variables are provided in Table S3 and Figure S2.

Main Results

Table 2 presents the results of DID analyses. In the whole sample, compared to non-DIP pilot hospitals, two mortality measures in pilot hospitals decreased after DIP payment, though not significantly. Both readmission measures decreased significantly by 3.2 percentage points (P <0.001) for all-cause readmission rate and 1.8 percentage points (P <0.001) for readmission rate with the same principal diagnosis. The rate of operation-associated complications or adverse events increased slightly but was not significant.

Table 2 Difference-in-Differences Estimates for the Quality of Care Associated with Diagnosis-Intervention Packet Payment

For heterogeneous effects across hospital levels, the results in columns (2)-(4) of Table 2 showed that the in-hospital mortality rate decreased significantly in tertiary hospitals by 0.6 percentage points (P <0.001) while increased significantly by 0.7 percentage points (P =0.022) in promoted secondary hospitals, and remained statistically unchanged in other secondary hospitals. The mortality rate of surgical patients and the rate of operation-associated complications or adverse events had no significant change at any hospital level, aligning with the whole sample. Both measures of readmission rates only decreased significantly in tertiary hospitals by 4.5 percentage points (P <0.001) and 2.1 percentage points (P =0.009), respectively, and decreased more than in secondary hospitals.

Table 3 shows the temporal dynamic quality changes after DIP reform. The in-hospital mortality rate, all-cause readmission rate, and readmission rate with the same principal diagnosis decreased significantly in the second year of DIP payment adoption by 0.3 percentage points (P =0.042), 5.9 percentage points (P <0.001) and 3.3 percentage points (P <0.001), respectively. The changes in the mortality rate of surgical patients and the rate of operation-associated complications or adverse events were not significant in either of the two years.

Table 3 Temporal Dynamic Changes in the Quality of Care After Diagnosis-Intervention Packet Payment Reform

As indicated in Figure 1, there were no pre-existing differential changes in outcome variables between the pilot and non-pilot hospitals prior to DIP reform. In the robustness checks (Table 4), the directions and significance of coefficients for all indicators remained consistent considering the outbreak of COVID-19, volume-based procurement of coronary stents, pilot of pay-for-performance, and the quality of discharge records.

Table 4 Robustness Checks Considering COVID-19 and Other Policies During the Study Period

Figure 1 Tests on the validity of parallel assumptions of quality measures. Abbreviation: DIP, Diagnosis-Intervention Packet. Notes: We replaced the post dummy in the interaction term in equation (1) with a series of year-month dummies. Changes in differences of the outcomes (A for in-hospital mortality, B for mortality of surgical patients, C for readmission within 30 days, D for readmission with the same principal diagnosis within 30 days, and E for operation-associated complications or adverse events) between patients in DIP pilot and non-DIP pilot hospitals relative to the first month of the study are plotted against the time to the DIP payment reform.

Additional Results

Two mortality measures in low-risk groups had no change in the whole sample and across all hospital levels, except for a significant decrease in the mortality rate of low-risk DIP groups in tertiary hospitals (Tables S4 and S5). Table S6 and Figure S3 presented a similar trend of CCI between the pilot and non-pilot hospitals in the overall sample and a larger increase in CCI after DIP payment in pilot tertiary hospitals compared to non-pilot tertiary hospitals. When exploring the heterogeneous associations with quality measures in different patient severities, we found that compared to patients with lower severity, the in-hospital mortality decreased significantly, and the two readmission measures decreased more in more severe patients, as shown in Table S7. Tables S8 and S9 and Figure S4 demonstrated a significant increase in the intensity of care measured by RW in the whole sample and tertiary hospitals. We also observed a generally more obvious decrease in quality measures in groups with higher RW (Figures S5–S9), especially for two readmission measures in the whole sample and tertiary hospitals. An exception is that for promoted secondary hospitals, the in-hospital mortality rate significantly increased in the group with a higher intensity of care.

Discussion

This study focused on the impact of the DIP payment reform in July 2019 on the quality of care in a large city in China. We found that DIP payment with a cap in the city was associated with improved healthcare quality. In particular, the readmission measures showed consistent results of both decreasing significantly, while the mortality measures and the quality safety indicator did not have significant changes. These results are robust considering other events or policies during the study period. Meanwhile, the quality improvement was primarily driven by tertiary hospitals and was more obvious over time after the policy adoption.

That the quality of care has been maintained, or even improved after case-based payment was consistent with several previous studies. For example, studies on in-hospital mortality in Poland,¹⁷ England,¹⁹ Japan,²⁰ and two cities^13,25 in China all showed no significant change after DRG or DRG-like payment. Another two studies on DIP payment in a different city in China also have similar results.^26,36 Regarding the readmission rate, while some studies reported an increase after the reform, such as patients in Switzerland¹⁵ and Japan,²⁰ many other studies found no change^{14,17,19,21,25,37} or even a significant decrease.²² The no change in the patient safety measure was consistent with a study in another DIP pilot city.⁶

There was no evidence of the selection of less severe patients since we did not observe an overall decrease in CCI in pilot hospitals, and the quality improvement was even more obvious in patients with higher severity, which was consistent with another study focused on the patient selection behavior after the DIP reform in this city.³⁸ Regarding the intensity of care, a previous study theoretically demonstrated that providers tend to increase the intensity of care for higher payments when providing different payments for different treatment intensities.¹² Our study empirically supported this inference by finding a significant overall increase in RW. Moreover, this behavior change partially contributed to the quality improvement, as quality measures in groups with higher intensity of care improved more. These two findings supported each other as sophisticated care would be generally provided to more severe patients. Another noteworthy point is that, despite the increase in RW, it would not necessarily result in an unexpected rise in total inpatient expenditure due to the cap. No change in the patient safety indicator could be attributed to the sustained focus of national and local health authorities prior to the payment system reform, leaving limited space for additional improvements.

The more obvious quality improvement in tertiary hospitals was consistent with our expectations and other studies. For example, a study from Korea found that the readmission rate in large hospitals declined more than in small hospitals after the DRG reform.²² A general explanation is that tertiary hospitals had a greater capacity to adapt and improve their clinical and administrative practices after the payment reform, which was also supported by our interviews with managers of pilot hospitals. Specifically, we found that tertiary pilot hospitals admitted more severe patients after DIP payment, meanwhile significantly increased the intensity of care. Moreover, the in-hospital mortality rate in the higher-intensity group exhibited a more obvious decrease than the lower-intensity group in tertiary hospitals, while the same indicator in the higher-intensity group, on the contrary, increased significantly in promoted secondary hospitals. Thus, similar hospital responses may not necessarily lead to similar quality changes, which may depend on hospitals’ clinical and administrative competence. Another possible explanation is that promoted secondary hospitals admitted patients with high-risk diseases after the reform, possibly transferred from tertiary hospitals. The evidence is that two mortality measures in low-risk groups did not have significant changes in secondary hospitals.

The more obvious quality improvement over time was also aligned with our expectations and what other studies have found already.⁶ At the beginning of policy implementation, providers had limited knowledge about the policy and its incentives and thus did not have immediate responses. As time went by, hospital administrators and physicians had a deeper understanding of the new policy through several training programs. Their knowledge was further reinforced after they received reimbursements, leading to stronger responses to the reform.

To our knowledge, this is the first empirical evidence of the comprehensive impacts of DIP payment on the quality of care and its underlying mechanisms. Using a range of quality measures, we found that the case-based payment with a cap applying a more granular classification system has the potential to improve quality, which was previously a major concern of policymakers and was thus in need of empirical evidence. We found that the quality improvement was at least partially attributed to the increase in intensity of care, especially in more severe patients. This major difference with the DRG payment may contribute to the broad roll-out of DIP payment in other cities in China and the payment reform in other low- and middle-income countries. However, we should keep in mind that quality improvement is not the main objective of case-based payment reform but a guarantee of its further roll-out. Long-term monitoring of the influences of DIP payment and its associations with specific diseases is also necessary. In addition, the differential impacts on tertiary and secondary hospitals need to be paid close attention in case the benefits of secondary hospitals and their patients are unevenly affected.

Limitations

Our study has several limitations. First, we did not explore the association of DIP payment with process measures of quality since the administrative discharge records we used did not contain further detailed information on the adherence to clinical guidelines and timeliness of care, etc., which could be more responsive to the policy reform. Second, we only observed the short-term influence in this city, and uncertainties remain in the long run and in other pilot cities. Although our study found a more obvious quality improvement during the first two years after the DIP payment, the quality outcomes and potential unintended results of the stronger provider responses in a longer period are unknown and need observation in the future. Third, due to data limitations, we did not observe the potential spill-over effects from EBMIS patients to other patients. Fourth, COVID-19 may have a lasting influence on hospital health services. However, we did not find a deterioration in the quality of care due to the increase in patient severity after the pandemic.

Conclusion

In this study evaluating the quality change after DIP payment in a large pilot city in China, DIP payment with a cap was associated with improved quality of care among patients in pilot hospitals. The provider’s behavior of increasing the intensity of care, especially for more severe patients, may partially contribute to the results. This may contribute to the further roll-out of DIP payment in other cities in China and the payment reform in other low- and middle-income countries.