Development of a novel hospital payment system – Big data diagnosis & intervention Packet

Highlights

• •The BD-DIP grouping is underpinned by real world data.
• •The BD-DIP is dynamic in grouping and reimbursement value.
• •The resource utilization within the BD-DIP groups is much more homogeneous.
• •Introduction of the BD-DIP showed budget savings and improvement in efficiency.

Abstract

The diagnosis related group (DRG) was the most commonly used prospective hospital payment platform in developed countries. One of the major limitations of the DRG system is that the DRG grouping is not sufficiently homogeneous in benchmarking underlying resource needs. We developed a novel hospital payment and management system called Big Data Diagnosis & Intervention Packet (BD-DIP) by applying the similar case mix index (CMI) principles but the grouping is based on unique combination of ICD-10 and ICD-9 v3 codes. The initial prototype of BD-DIP was developed using hospital discharge records in Shanghai and then piloted in Guangzhou, China. The average coefficient of variation of the DB-DIP is about one-third smaller than the US DRG system. Results from the pilot evaluation showed that introduction of the BD-DIP lead to about 5% hospital budget savings and notable improvement in hospital care efficiency, including increased institutional CMI, lower admission rates, smaller variation in hospital charges, and lower patient cost-sharing burdens. The implementation of hospital monitoring tools resulted in identification of potential irregular practices to enable further auditing and investigation. The BD-DIP platform has a number of advantages over DRG-based payment models in terms of more homogeneous resource utilization within groups, design simplicity, dynamic in grouping, and reimbursement value in reflecting real-world treatment pathways and costs, and easy to implement.

1. Introduction

The last two decades have witnessed continued rise in expenditures on healthcare worldwide. Between 2000 and 2017, global health spending in real terms grew by 3.9% a year while global GDP grew 3.0% [1]. The growth in healthcare spending was even faster in low and middle income countries, rising 7.8% and 6.0% each year, respectively, during the same period. In 2017, globe spending on healthcare amounted to about 10% of GDP. Hospitals generally accounted for the highest proportion of healthcare expenditure. Among EU member states and in the US, hospital care cost made up 36.3 % and 32.7% of healthcare expenses, respectively, in 2017 [2], [3]. Constrained by increased hospital care needs and limited resources, healthcare policy makers and payers have been on constant pursue for hospital cost containment. Towards this effort, a patient classification system, referred to as the diagnosis related group (DRG), has been increasingly used in many countries as a tool for both measuring and benchmarking hospital activities and their funding [4].

The DRG system was first introduced in the US in 1983 in response to rapidly increasing healthcare costs by replacing retrospective payments with prospective payments for hospital charges [5]. Under a prospective system, hospitals are paid a set fee for treating patients in a single DRG category regardless of the actual cost of care accrued. Encouraged by initial reduction in hospital activities following introduction of the DRG [6], many countries started to implement their DRG systems by either direct coping or developed DRG variants to suit the local context, needs, and capacities [7], [8], [9].

The DRG systems in different countries are in a state of continuous evolving. One of the common trends in this regard is that the number of groups has been expanded considerably in recent years, likely reflecting the number of DRG groups introduced in the earlier years not being sufficiently homogeneous in benchmarking underlying resource needs. For examples, the number of DRG groups in Germany increased by nearly 40% between 2005 and 2011; the number in England more than doubled, and the number in France more than tripled [7]. The impact of DRG-based prospective payment system on hospital expenditures has been mixed. While in the US, DRG based payment helped contain costs, most countries in Europe observed an increase in hospital activities leading to higher hospital costs upon implementing DRG-based payment [7]. Unlike US, where DRG based payment succeeded fee for service, most countries in Europe moved to DRG based payment from global budgets. DRG based hospital payments may also have other unintended consequences by encouraging opportunistic practices. To enhance profit, hospitals may discharge patients early and then readmit them; increase case volumes by reducing admission criteria; or up-code a patient in a DRG with a higher reimbursement rate, etc [7], [10], [11], [12]. These erratic and sometimes harmful behaviors run counter to the rationales for the DRG-based payment system and neutralize its desired effects.

In China, the prevailing method of payment for public hospitals is regional-based government reimbursement of hospitals on a fee for service basis with a disease-specific cap for each admission. Consequently, over-treatment and over-prescription caused by the fee for service payment system were widespread in China [13]. Between 2007 and 2012, health expenditure grew by 14.9% annually, outpacing the GDP growth by nearly 5% [14]. China had also experimented DRGs as early as the beginning of the century at the regional level [15]. The province of Jiangsu began its piloting phase for a simplified DRG system in 2001 and this phase terminated in 2008 [16]. Beijing launched its local developed DRG after 5 years research without success in practice with just 8 hospitals participating [17]. Chengdu (capital of Chongqing province) initiated a feasibility research of introducing Australian refined DRGs but ended without expanded use [18]. In Shanghai, local health authority piloted a prospective payment system in 2004 whereby a fixed fee was imposed on selected diseases [19]. Despite support from regional government, the effects of the various forms of DRG-based prospective payment have not been successful and a fully functioning DRG system has remained unavailing in China today.

The initial DRG platform was envisaged based on then primitive healthcare IT system. Despite numerous upgrades over time, DRG-based systems have been proven to be complex and rationales for the grouping, weight assignment and the reasons for revision rely on input from medical specialists and expert consultants. The processes tend to be opaque and time consuming, and are hard-pressed to meet the needs of today’s explosion of healthcare technology innovation. The essential foundation of a DRG system is built on the case mix index (CMI), which is used to group patients with similar clinical characteristics and relatively homogeneous resource consumption with a DRG weight assigned to each group based on treatment cost. In the context of the challenges and limitations for implementing DRG-based hospital payment system in China, we have designed a novel CMI system which is derived directly from real-world inpatient care data, hereinafter referred as Big Data Diagnosis & Intervention Packet (BD-DIP). This novel hospital payment system is built upon the previous framework by Yang and Reinke [20], who have shown that using ICD-derived CMI is a valid alternative as a proxy of severity of illness and as a measurement of confounding for hospital-based health service research in countries that do not have the DRGs.

In this communication, we describe the key elements and their rationales in developing the BD-DIP system both as a benchmark for hospital payment and hospital management tools for monitoring irregularities of patient care and share our initial experiences in piloting this platform.

2. Methods

The prototype platform was initially conceptualized and developed based on acute care hospital discharge records from 2014 to 2017 in Shanghai and the methodology was subsequently updated and pilot tested in Guangzhou of China in 2018 and 2019. Over the past decade, public hospitals in China have had substantial investment in healthcare IT system, including standardized hospital discharge records. Among others, mandatory fields included discharge diagnoses and procedures performed coded with ICD-10 and ICD-9 v3, respectively. As all personal identifiers were deleted during data extraction, no ethical committee review is required per local policy on use of electronic health information.

2.1. Overview of the BD-DIP platform

The BD-DIP platform is composed of two distinct and complement components; DIP Grouping Database and Supplement Catalogs (Fig. 1). The DIP Grouping Database is the novel inpatient care classification system containing 3 levels of hierarchal group listings and a Master Index. Supplement Catalogs support hospital reimbursement by applying the payment adjusters (Resource Utilization Calibrators) and providing evidence-based tools for hospital behavior overseeing.

Fig. 1.

Components of BD-DIP Platform. BI: The Balance Index; RRA: Rating of Repeated Admissions; RLA: Rating of Low- RW Admissions; REH: Rating of Extended-stay Hospitalization.

2.2. DIP grouping Database

Fig. 2 provides an example of different levels of listings for vascular heart disease. The building blocks for BD-DIP platform are unique combinations of disease and intervention groups, which constitute the Level 3 listings for hospital payment. However, unlike the DRG system, the grouping criteria of BD-DIP are determined solely by ICD-10 codes (first 4 digits) in combination with ICD-9 v3 codes as recorded in hospital discharge records. In the initial design of BD-DIP prototype, a core DIP group was formed if the number of cases within a given DIP exceeded 15 patients and in so doing, >85% of total discharge records would be included in the core DIPs. The remaining DIPs are considered mixed and aggregated under ICD-10 codes and 4 intervention modalities: Conservative Treatment, i.e., medications, Diagnostic Procedure, Therapeutic Procedure, and Surgical Operation. There are 3 subgroups under therapeutic procedure and 3 subgroups under surgical operation based on anticipated resource needs.

Fig. 2.

An example of level of listings for vascular heart disease.

The Level 3 listings are consolidated into the Level 2 listings taking into consideration of resource requirement for available interventions and complexity of therapies. The Level 2 listings are intended for use of hospital management, including budgeting and performance evaluation. While there are multiple intervention modalities for a given diagnosis, the guidance of Level 2 listings is that costs for inpatient services within the group are similar. The Level 1 listings are built upon the Level 2 under the grouping of disease types regardless of underlying intervention costs. The Level 1 listings are further organized by a Master Index based on anatomic and/or etiology classification similar to the ICD system. The Level 1 listings in conjunction with the Master Index are designed for hospital service planning from perspective of healthcare delivery.

To compare homogeneity of resource consumption within groups classified by the BD-DIP and US DRG grouping methodologies, we estimated both coefficient of variation (CV) and reduction of variance (RIV) using the discharge records from Shanghai in 2018.

2.3. Supplement Catalogs - resource utilization Calibrators

Similar to the DRG system, we developed a number of Resource Utilization Calibrators (RUC) to more accurately reflect underlying medical resource needs based on analysis of expenditures of various subpopulations, including Disease Severity Calibrator, Cancer Staging/Tx Calibrator, 2nd Diagnosis Calibrator, and Age Calibrator. All coefficients were derived based on the following formulae.

$$\mathrm{R}\mathrm{U}\mathrm{C}={\mathit{r}\mathit{w}}_{\mathit{j}}/{\mathit{r}\mathit{w}}_{\mathit{i}}$$

Where ${\mathit{r}\mathit{w}}_{\mathit{j}}$ is the related weight for the subpopulation $\mathit{j}$ within an aggregated DIP group derived using the same method for ${\mathit{r}\mathit{w}}_{\mathit{i}}$ estimate as described below. For example, ${\mathit{r}\mathit{w}}_{\mathit{j}}$ may refer to patients with metastatic esophageal cancer receiving radiation treatment in addition to chemotherapy and ${\mathit{r}\mathit{w}}_{\mathit{i}}$ in this case refers to the average DIP value for patients with metastatic esophageal cancer receiving chemotherapy.

2.4. Development of DIP values and case mix Index for hospital payment

2.4.1. Related weight (${\mathit{rw}}_i$) for a given DIP

Under the current DRG payment system, each DRG is assigned a related weight based on the average amount of resources that it takes to care for a patient assigned to that DRG. Hospitals are paid for each inpatient care by multiplication of the base rate and the related weight of a given DRG. The BD-DIP employs the same approach in estimating the related weight of each DIP but with the cost data being sourced from real world settings. Briefly, ${\mathit{rw}}_i$ for a given DIP is expressed as follows.

$${\mathit{rw}}_i=m_i/M\mathrm{X}1,000$$

Where m_i is the average cost for a given DIP group i and M is the average cost for the reference DIP. We chose laparoscopic appendectomy as the reference DIP, i.e., ${\mathit{rw}}_i$ value for a given DIP group was determined relative to the average cost for laparoscopic appendectomy, which had a default value of 1,000. Table 1 illustrates three DIP groups for treatment of appendicitis and their corresponding ${\mathit{rw}}_i$ values. Any other DIP’s ${\mathit{rw}}_i$ values are determined accordingly.

Table 1.

Examples of DIP Groups for Treatment of Appendicitis and their Corresponding ${\mathit{rw}}_i$ Values.

ICD-10 (K35.9)	Procedure (ICD-9 v3)	Average Cost	${\mathit{r}\mathit{w}}_{\mathit{i}}$
Appendicitis	Conservative treatment (N/A)	¥4,099	306
Appendicitis	Other appendectomy (47.09)	¥9,173	686
Appendicitis	Laparoscopic appendectomy (47.01)	¥13,385	1,000

2.4.2. Institutional overall related weight (RW), payment Rate, and case mix Index (CMI)

Each hospital’s total RW is determined by aggregation of all possible DIP ${\mathit{rw}}_i$ values in that hospital multiplied by number of cases in the corresponding DIP expressed as follows.

$$\mathit{RW}=\sum _{i=1}^k{\mathit{rw}}_i\times n_i$$

Where ${\mathit{rw}}_i$ is DIP value for group $i$; $n_i$ presents number of cases in DIP group $i;$ and k presents the number of DIP groups in a given hospital.

All hospitals’ RW within a geographic area will be aggregated together to obtain the total RW within entire catchment area. The total available reimbursement fund in the area will then be divided by the total RW for all hospitals within entire catchment area to arrive the basic payment rate of each rw. If no payment adjustment is considered, the amount of any given hospital payment received would be equal to that hospital total RW times the basic payment rate of each rw calculated above.

To further support appropriate payment adjustment, institutional average CMI is estimated as follows, which reflects the overall patient disease severity and treatment complexity at a given hospital.

$$\mathit{CMI}=\frac{\mathit{RW}}{{\sum }_{i=1}^k{n}_i}÷1,000$$

The calculation of institutional $\mathit{RW}$ and $\mathit{CMI}$ is illustrated in Fig. 3.

Fig. 3.

Estimate of Institutional Overall Related Weight and Case Mix Index.

2.5. Supplement Catalogs - tools for monitoring hospital irregular behaviors

Experiences from implementation of DRG payment system have revealed that hospitals may engage certain erratic inpatient care practices to enhance their revenue, including up-coding of billing code, separate hospitalization and re-admission of patients, lowering admission criteria, etc. and we have designed a number of tools to oversee and discourage system-gaming practices.

2.5.1. The Balance Index (BI)

As RWi values for DIPs serve as adjusting factors for hospital payment, an institute may have incentives to perform one particular DIP (RWi) service but submit the claims with the same diagnosis but slightly different interventions with higher RWi, or just to opt for the higher RWi interventions to enhance its profitability.

Assume we are interested in inpatient care behaviors for a hospital for a given diagnosis which has k DIPs (e.g., metastatic breast cancer) relative to the benchmark, i.e., inpatient care for the same diagnosis in a defined catchment area. We first rank RW_i for each DIP from the lowest to highest value. The number of cases in the catchment area for the corresponding RWi is n_i, where i = 1, 2, 3……k. We further assume that we are interested in RWi value greater than j, i.e., the cut-off value divides k DIPs into two groups, lower RWi and higher RWi, where the lower RWi value group includes DIPs from 1 ∼ j and the higher RWi value group includes DIPs from j + 1 ∼ k.

The average RW for the lower RWi value group is calculated as ${\mathit{RW}}_l=\frac{{\sum }_1^j({\mathit{rw}}_i\times n_i)}{{\sum }_1^j{n}_i}$

The average RW for the higher RWi value group is calculated as ${\mathit{RW}}_{\mathrm{h}}=\frac{{\sum }_{j+1}^k({\mathit{rw}}_i\times n_i)}{{\sum }_{j+1}^k{n}_i}$

The Balance Index is calculated as

$$\mathrm{B}\mathrm{I}=\frac{{\mathit{SRW}}_{\mathrm{h}}/S{\mathit{RW}}_l}{{\mathit{RW}}_{\mathrm{h}}/{\mathit{RW}}_l}$$

Where ${\mathit{SRW}}_{\mathrm{h}}/S{\mathit{RW}}_l$ represents odds (higher vs. lower cost DIPs) calculated for a given hospital using the same approach as described above. And RW_h/RW_l represents pooled average odds in a defined geographic area. When ratio of the two odds (BI) is equal to 1, it implies that inpatient care for the DIPs of interest at this institute is identical to the benchmark of entire catchment area. When the BI is>1, it implies that this institute has higher RW values compared to the benchmark and serves as a red-flag for potential up-coding of its DIPs. In practice, consideration should be given for the appropriate benchmark, e.g., all hospitals within the same tier level and the findings should be viewed in the context of similar data prior to implementing the BD-DIP. With a pre-set cut-off value of BI, the hospitals irregular behaviors can be monitored in the systematic fashion for identification of potential offenders which may warrant further investigations.

2.5.2. Rating of Repeated Admissions (RRA)

Repeated admissions imply that patients are re-hospitalized within 30 days ensuing discharge from the same hospital for the same diagnosis. We rate the hospital re-admission in 5 categories in terms of its deviation from the benchmark using the standard deviation with weight applied for calculating hospital-wide re-admission rating. Hospitals with significantly high RRA are subject to further investigations.

2.6. Rating of Low- RW Admissions (RLA)

The purpose of rating of low- RW Admissions (RLA) is to monitor and discourage high admissions of patients with illnesses that can be mostly managed at ambulatory settings. We first establish the benchmark by calculating system-wide admission rates and their standard deviations for conditions that should be treated at outpatient settings for most patients, e.g., benign oral tumor. Hospitals substantially deviating from the benchmark, e.g., >=2 or 3 standard deviations for majority of the conditions could be flagged for auditing.

2.7. Rating of Extended-stay hospitalization (REH)

Under the prospective payment system, there is little incentive to unnecessarily extend the patients’ stay in hospitals. Nevertheless, REH is an important index to evaluate hospital performance efficiency and for health care system planning. The benchmark is established by calculating mean hospital stay for each DIP. We flagged percentage of patients with hospital length of stay >=2 times longer than the benchmark in estimating the REH.

3. Results

3.1. The prototype of BD-DIP platform vs US DRG system

Approximately 3.7 million hospital discharge records involving financial transactions in the amount of 63.8 billion Yuan (∼9 billion US $) in 2018 from Shanghai were analyzed and compared using the US DRG and BD-DIP platform separately. The US DRG methodology led to 94.7% of the discharge records being grouped into 771 DRG categories with financial transactions within the DRG groups amounting to 87.2% of total financial transactions. The BD-DIP classification resulted in 98.6% of the patients being grouped into approximate 14,000 core DIP and 2,499 mixed DIP groups (Level 3) with financial transactions within the DIP groups amounting to 95.4% of total financial transactions. Compared to the BD-DIP platform, the DRG methodology brought about a larger amount of expenditures not being included in the standard DRG groups, suggesting that patients requiring disproportionally higher resource utilization tended to not being handled by DRG payment schemes.

The hierarchical-structured prototype of BD-DIP system also had 129 listings in the Master Index, 1,194 listings at the Level 1, and ∼ 3,000 listings at the Level 2. The average CV at the Level 3 listing for the BD-DIP platform was 0.6383, which was 34.5% lower than the average CV if the DRG classification was employed (0.9743), indicating greater homogeneity in healthcare expenditures within DIP groups relative to the DRG groups. The RIV values, on the other hand, were quite similar with 0.3501 for the BD-DIP and 0.3433 for the DRG platforms, respectively, indicating similar heterogeneity between the groups.

3.2. Early experience from piloting the BD-DIP payment platform

The prototype BD-DIP payment platform was subsequently piloted in 2018 and 2019 in Guangzhou, which had approximate 15 million permanent residents and close to 100% were covered by government sponsored medical insurance. Table 2 displays major matrix of the BD-DIP platform in processing hospital payment over the testing period. In the pilot testing, a core DIP was formed when the number of patients within this DIP was 10 or more to ensure approximate 95% of all discharge records were included in the core DIPs. The classification of mixed DIP was also streamlined to further enhance the automation of patient grouping classification. In 2018, 95.1% of discharge records were included in 9,972 core DIPs and 25 mixed DIPs. In 2019, additional 1,709 new core DIPs were added and 1,029 existing core DIPs were removed, resulting in 94.8% of discharge records being included in 10,652 core DIPs and 25 mixed DIPs.

Table 2.

Matrix of BD-DIP Platform for Hospital Payment in 2018 and 2019 in Guangzhou.

Characteristics	2018	2019
Number of Participating Hospitals	334	338
Number of Hospital Discharge Records	1,207,000	1,352,800
% of Insured Population Receiving Hospital Care (Previous Year -> Current Year)	15.2% -> 14.3%	14.3 -> 13.7%
Number of Core DIPs	9,972	10,652
Number of Mixed DIPs	25	25
${\mathit{rw}}_i$ Distribution (%)  < 1,000  1,000 – 2,000  > 2,000	73.8% 18.1% 8.1%	66.5% 19.5% 14.0%
Change in CMI vs. Immediate Preceding Year	+ 2.10%	+ 3.76%

In 2018, the first year for piloting the BD-DIP payment system, the RUC was not implemented nor the tools to monitor hospital irregular behaviors. In making reimbursement to hospitals, we made a number of simple adjustments for institutional total RW so hospitals with disproportionally higher number of patients requiring greater medical resources for patient care were appropriately reimbursed. For institutes with CMI > 1 (an indicator for above situation), there was an upper adjustment of 1% over base payment for each 0.1 increment of CMI capping at 10% over the base payment threshold (RW value). For institutes with a higher proportion of elderly patients, there was an upper adjustment of 1% for each 10% increment of senior population capping at 5%. Hospitals rated AAA and AA levels were rewarded additional 1.5% and 0.5% RW, respectively. Finally, institutes with each designated specialty care service were given extra 0.5% RW. Guangzhou overall hospital budget was approved at ¥12.8 billion and the total hospital reimbursement was ¥12.1 billion (94.7% of the budget), of which ¥11.7 billion was to reimburse hospitals for inpatient care services and ¥0.4 billion was paid to 141 institutes (42.2% of participating hospitals) as performance bonus.

For the fiscal year of 2019, adjudication of payments to some hospitals has not been finalized as of writing this manuscript because of ongoing investigations of suspected irregular billings. A total of ¥0.77 billion billings were flagged as potential up-coding of DIPs accounting for 3.7% of total billing amount. Through analysis of Rating of Low- RW Admissions, about 52,000 patients involving billings of ¥0.21 billion were considered admitting to hospitals inappropriately. In addition, 6767 patients involving billings of ¥0.12 billion were found to have repeated admissions for the same condition within 7 days of discharge from the same hospital or same tiered hospitals.

In 2018 and 2019, system-wide overall CMI was increased by 2.10% and 3.76%, respectively. However, hospital admission rate was reduced by 0.9% and 0.6% vs. immediate preceding year (15.2%, 14.3%, and 13.7% for 2017, 2018, and 2019, respectively). The average patient out-of-pocket payment for each admission was ¥7,034, which was 2.4% lower compared with the year prior to the piloting.

4. Discussion

Ideally, a hospital reimbursement system should have benchmark that accurately reflects the underlying cost of individual patient care and has built-in mechanisms to monitor and deter system gaming or manipulation. With this in mind, our goal is to develop a ground-breaking hospital payment system with existing IT infrastructure and data sources that can more accurately represent resource needs and is transparent, objective, and adaptable. The preliminary findings from our pilot testing are encouraging. The introduction of the BD-DIP had no negative impact on hospital budget and resulted in notable improvement in hospital care efficiency, including increased institutional CMI, lower admission rates, smaller variation in hospital charges for the same DIP groups, and lower patient cost-sharing burdens.

Similar to the DRG system, the BD-DIP platform applies the same principles of methodology of CMI as a benchmark for hospital payment. In essence, a hospital's DRG measures the complexity of cases treated at that particular hospital relative to the average complexity in a peer group of hospitals [21]. As hospital CMI is usually derived from the DRG and reflects aggregate risk of individual patients within a hospital [22], CMI measures the intensity of services and consumption of resources for each DRG.

The BD-DIP platform has a number of unique features which make it an attractive alternative for hospital payment. In designing US Medicare DRGs, patients were characterized and clustered based on expected similar resource consumption. However, largely owing to limited number of DRG groups, concerns were raised that earlier DRG system does not fully reflect resource-intensity differences in severity of illness and inadequacy remains even with the market-basket payment adjustment [23]. In addressing this limitation, Medicare payment model has been continuously updated over time, including the DRG codes assigned to different constellations of ICD-9 diagnoses and procedures, the weights assigned to those DRGs, and the specific methodology for geographic adjustments [24]. DRG systems implemented in EU countries have also been criticized for inadequate explanatory power of resource utilization. It has been shown that a standard set of patient characteristics and treatment variables explain the variation in costs or length of stay better than the DRG variables at least for certain medical conditions [25]. The uneven resource-intensity within the DRG groups may financially motivate hospitals to engage in unintended behavioral responses, such as early discharge, skimping, cream-skimming or dumping, and upcoding [26], [27], [28], [29], [30]. It has been proposed that DRG-based systems need to be able to define sufficiently resource-intensity homogenous groups to mediate these potential unintended consequences and to serve as the basis for fair hospital payment and performance comparisons [7]. Towards this goal, one of the most significant advantages of the BD-DIP system is that resource-intensity within DIP groups is much more homogeneous than DRG-based systems as we have shown that the BD-DIP had considerably smaller CV compared to current US DRG when the same hospital discharge records were analyzed. Furthermore, the BD-DIP platform is dynamic by design; the number of core DIP groups is not fixed but varies reflecting current patient volume of case mix. Similarly, the value for DIP groups as determined by ${\mathit{rw}}_i$ is updated with real-world data in real time, reflecting standard of care and adoption of latest healthcare technology. Howerver, as some DIPs involve relatively small number of cases, heterogeneity of samples can mean that ${\mathit{rw}}_i$ varies significantly from one year to the next. From a practical point of view, in deriving ${\mathit{rw}}_i,$ the healthcare administrators may choose administrative claims data of multiple years prior with heavier weights applied to most recent years for more robust and accurate estimates of resource weight. As expected, more homogenous resource utilization within DIP groups was achieved at the expense of increased case mix groups, which are characterized by a distribution patter of highly skewed to the right with a long tail. From hospital management perspective, selected DIP groups with high patient volumes which consume disproportionally high resources should be primary focus. Unlike the DRG system, the BD-DIP system is much more objective and transparent for both payers and payees. Finally, the BD-DIP platform is built on existing administrative claims database. As such, there is no new coding required for the core BD-DIP groups and can be implemented with minimum burden to hospitals. In designing the mixed DIP groups, there is a trade-off between more homogeneous resource utilization within the groups and less efficient data automation. Although the prototype of BD-DIP system had 4999 mixed DIPs, we decided to coalesce mixed DIP groups into 25 groups in our pilot testing for more efficient data automation with existing hospital IT system. With increased use of AI technology in healthcare IT, re-coding for mixed DIP groups can be expected to accomplish much more efficiently in future. The BD-DIP system is more appealing for developing countries as development of a new DRG-based system is complex and countries must first establish the infrastructure, human resource capacity, information management systems, etc.

The BD-DIP system has a number of built-in tools to monitor and deter the system-gaming through up-coding and patient selection. We did not deploy these tools but opted to employ a number of simple payment adjusters in the first year of piloting because it is important to seek buy-ins from stakeholders and also hospitals are likely to engage in certain behaviors for financial gains once they know how the new system works. All monitoring tools were developed and benchmarked by flagging high-cost outliers that exceed a predetermined threshold outside of expected “normal” boundary. It is important to point out that variation in hospital care should be expected and healthcare administrators should set up their own benchmark for auditing after discussion with hospital stakeholders taking into consideration of the local practices. They may selectively choose certain tools or develop their own monitoring tools over time in addressing local unique situations. Similarly, the RUC is not cast in stone. For example, local population characteristics, priority for disease prevention, or preferred intervention strategies could all be part of consideration in designing incentives for hospital payment adjustment. Indeed, we have just initiated a new pilot testing on inpatient care of five common medical conditions (lung cancer, colorectal cancer, breast cancer, atrial fibrillation, and acute appendicitis). The hospitals will be rewarded with additional payment if they can meet predetermined performance targets, i.e., greater utilization of preferred interventions guided by evidence-based medicine leading to better clinical and economic outcomes. Through discussion with hospital stakeholders and implementation of the BD-DIP, we observed some positive changes for hospital patient care. CMI scores were increased in both 2018 and 2019 accompanied by consecutive decrease in percentage of covered population receiving inpatient care, suggesting shifting of inpatient care to outpatient settings for less severe conditions. We also observed smaller variation in healthcare resource consumption. Prior to implementing the BD-DIP, 24.1% of the hospital payments were outside the boundary of 50% to 200% of the average payment for a given DIP and payments outside of this boundary were reduced to 20.6% by 2019. It is notable that the findings from pilot testing also signified satisfactory feasibility and plausibility of the BD-DIP platform for its potential to be implemented across the country as the piloting was conducted in a major metropolitan of different region instead of the same area from which the prototype of BD-DIP system was developed. Encouraged by the early experience from the piloting, the National Healthcare Security Administration of China has officially launched the next wave of expanded testing of BD-DIP system by including 71 cities in 21 provinces [31].

The BD-DIP system has a number of limitations. Core DIP groups are created if the number of cases for a given diagnosis and intervention combination exceeds a pre-determined threshold, which is arbitrary and also has a trade-off. The fewer number of cases required for a core DIP group would lead to greater number of core DIP groups. However, the estimated ${\mathit{rw}}_i$ may be less robust and subject to potential outliers linking to payments. Conversely, a higher threshold for formation of core DIPs would lead to more patients not being included in the core DIPs and assigning these cases into mixed DIPs or handling them on a case-by-case basis requires more work and also being viewed as more subjective and opaque. Before the BD-DIP payment platform can be implemented, there should be a process established to pay for patients in mixed DIPs, which should suit the local context and needs akin to the outlier payment arrangements in other activity based funding systems. The BD-DIP platform only functions as it intends if the underlying data source is accurate and reliable. For hospitals used to receiving finance through fee for services or global funding, there is little incentive for hospitals to capture all relevant data required for payment adjustment. However, it can be expected that hospitals would make efforts to improve coding depth and quality and to ensure that the data adequately reflect the complexity of patients treated following introduction of the new payment model [32]. The present study only described the design features of a novel hospital payment system and initial experiences through piloting, we did not examine its impact on quality of care, e.g., in-hospital mortality and patient-centered outcomes. It is important to include clinical outcomes in future studies to ensure that introduction of a new payment system does not negatively affect patient outcomes.

5. Conclusions

The BD-DIP platform is a novel system for hospital payment and performance evaluation based on existing hospital IT infrastructure and data sources. It is based on the same CMI methodology for which DRG system is developed but has a number of advantages over DRG-based payment models; more homogeneous resource consumption within groups, design simplicity, dynamic in grouping, reimbursement value in reflecting real-world treatment pathways and costs, and easy to implement. It is particularly appealing in countries that currently do not have the DRGs.

6. Credit authors’ statement

JX and SX were co-leads of concept and design of the BD-DIP and the piloting. HX, XC, XY, and XH also made contributions to the development of BD-DIP, including interpretation of data for the work. XC and JX had responsibilities for data acquisition and analysis. XH and JX led on drafting the manuscript and other authors contributed in revising the manuscript. All authors approved the final version of manuscript and agreed to be accountable for all aspects of the work.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.