Development of data-driven clinical pathways: the big data clinical evidence-based pathways

Abstract

Objectives

This study developed clinical evidence-based pathways (CEBPWs) to standardise treatment protocols, align diagnosis-reimbursement criteria, detect upcoding and enable early overtreatment warnings.

Methods

The CEBPWs were developed based on hospitalised patient-level data from 1 January 2022 to 31 June 2024 in 166 public hospitals in 16 administrative districts of Shanghai. It includes a total of 5 319 336 cases, involving 3 688 108 groups of ‘diagnosis+therapy’. 2.61 billion records of hospitalisation charges and 876.45 million records of outpatient charges were collected. GROWTH algorithm was used to find the combination of frequently charged items for examination, treatment, drugs and devices in ‘diagnosis+therapy’ group.

Results

CEBPWs comprise five key elements: objective evidence identification, accurate classification, value weighting, frequency weighting and temporal sequencing of evidence. We applied CEBPWs to 224 diseases, detecting issues including upcoding, overtreatment and fragmented care episodes to enhance healthcare quality. CEBPWs achieve 100% coverage in diagnostics, therapy and consumables, with 81.81% drug coverage. The pilot evaluation showed that there were violations in 433 cases, with a frequency deviation of 8.64% and cost deviation of 10.82%, with 8.95% for diagnosis, 9.44% for therapy, 14.81% for drugs and 8.98% for consumables.

Discussion

We were developed CEBPWs, breaking the limitations of the clinical pathways is that the experience of clinical experts rather than objective criterion based on the characteristics of big data and lack of diagnostic and therapy standards integrated with payment standards.

Conclusion

The results indicate that CEBPW is critical tool for hospital management and regulation, address many drawbacks of clinical pathways.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• There were some previous studies that regulated medical behaviour using clinical pathways, which the experience of clinical experts rather than objective criterion based on the characteristics of big data. However, no study attempted to develop clinical evidence-based pathways (CEBPWs) using big data.

WHAT THIS STUDY ADDS

• In this study, which included 4.7 million data front sheet of the medical record, the digital technology-based FP-GROWTH algorithm that CEBPWs found violations in 433 cases, reflects the data characteristics of the clinic and is more objective, realised the unity of diagnosis standards, therapy standards and payment standards.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• The CEBPWs can accurately detect abnormal case of upcoding, overtreatment and unreasonable treatment using clinical big data and FP-GROWTH algorithm, it holds potential for future utilisation in providing hospital management, and medical insurance reform and supervision.

Introduction

According to the 2010 WHO report, the key characteristics of good health service delivery include improvements in both the efficiency and quality of healthcare.¹ Optimising the quality of healthcare services, regulating medical behaviour, reducing costs and innovating hospital management and operations while balancing the relationship between the price and quality of medical services remain the primary focus for many countries in the healthcare sector.^{2 3} Clinical pathways (CPWs),^{4 5} the diagnosis-related group⁶ and Big Data Diagnosis & Intervention Packets (BD-DIPs)⁷ are important tools for hospital payment and management.

Originating in the USA, CPWs have been used in healthcare since the 1980s. The widespread use and prevalence of CPWs in hospitals is reported in the USA, Australia, Canada, Europe and Asia.⁸ However, with the advent of digitalisation and information, CPWs have also shown many limitations and shortcomings. First, CPWs are subjective criteria based on the practical experience of clinical experts and manual methods rather than objective criteria based on the characteristics of big data of patients. In the era of medical science and technology and precision therapy, a huge number of disease combination clusters and variables are generated, whereby the formulation of CPWs exceeds the scope of manual ability and thus entails a huge, low-efficiency, long-term and high-cost project.9,11 Second, a novel hospital payment system and standard, BD-DIPs,⁷ has been developed and future reform of payments and hospital management will rely more on real-world data analysis, yet diagnostic and therapeutic standards based on big data have not been developed. Therefore, under the premise of DIP payment reform, CPWs have not yet been deeply integrated with DIPs to maximise their effectiveness. Third, because CPWs are a set of standardised treatment modes, the number of cases entering clinical pathway management is low and CPWs have not been sufficiently implemented in hospitals,¹² which generates variability and bias that do not allow them to be adapted to individual patient needs.^{13 14} Four, CPWs cannot use digital intelligence systems to provide visual presentations and quickly offer intuitive recommendations to hospitals, departments and doctors. Finally, the incentive mechanism is imperfect and medical staff are not highly motivated.¹⁵

The above analysis shows that CPWs provide diagnostic and therapy standards, but it can be seen that digital technology has subverted the traditional CPWs, which do not fully reflect the real-world overview, through big data characteristics. Payment standards based on DIPs have been established, but diagnosis and therapy standards based on big data have not yet been established to integrate and support DIPs. Our research question is how to use big data to establish diagnostic and therapy standards, objectively reflect the path of diagnosis and therapy in the real-world data, identify the relationship between different combinations of multiple variables, support them together through medical insurance payment standards and provide objective evidence for hospital management and medical insurance payment.

Objective

This study aims to develop a novel type of clinical pathway that we call clinical evidence-based pathways. We describe the key elements of CEBPWs and their rationales and also share our initial experiences in piloting this platform. CEBPWs are both diagnostic and therapy standards and hospital management tools for monitoring irregularities in patient care. CEBPWs and insurance payment support each other, forming a closed-loop management of medical process standardisation, cost control and medical insurance payment.

Method

Data source

The CEBPWs were initially conceptualised and developed based on hospitalised patient-level data from 1 January 2022 to 31 June 2024 in 166 public hospitals in 16 administrative districts of Shanghai, including 56 tertiary and above hospitals and 110 secondary and above hospitals. CEBPWs were created using the original ‘diagnosis+therapy’ data, it includes a total of 5 319 336 cases, involving 3 688 108 groups of ‘diagnosis+therapy’. The data comprise three parts: first, a total of 5.31 million data points were collected from the front sheets of the medical records, including the disease diagnosis information, of hospitalised patients, and the therapy methods and costs during hospitalisation; second, 2.61 billion records of hospitalisation charges were collected, including charged time, examination items, therapy methods and fees during hospitalisation; and third, 876.45 million records of outpatient charges were collected, about 856.94 million cases. As all personal identifiers were deleted during data extraction, no ethical committee review is required as per local policy on the use of electronic health information.

Data processing and statistical analysis workflow

First, the researchers combined the database of admission records, records of hospitalisation charges and records of outpatient charges using unique identification numbers, and cleaned up the data. The items were categorised and labelled as diagnosis, therapy, drugs and consumables, the original charge data of patients were reduced in dimension, and the main features of the data were retained.

Second, according to the classification rules of ‘diagnosis+therapy’ group. The total cost of each label for each disease was calculated, they were arranged in descending order, and the cumulative sum was calculated from top to bottom. Of these items, 80% were intercepted and the rest were eliminated. According to statistics, 20% of all cases accounted for 80% of the costs.

Third, the FP-GROWTH algorithm was used to find the combinations of frequently charged items for examination, treatment, drugs and devices in each ‘diagnosis+therapy’ group. Using the arules package eclat function (which implements the FP-growth algorithm), with a parameter support of 0.7, we constructed FP-Tree and mined the frequent item sets in FP-Tree, then looped out a portfolio of frequently charged items in different categories such as drugs, treatments and consumables under the DIP directory. Items that 100% of patients undergo are called mandatory items, items that 95% of patients undergo are called high-frequency items, and items that 30% of patients undergo are called low-frequency items. Mandatory and high-frequency items are based on common data characteristics, and these items will impact medical quality, low-frequency items are determined based on patients’ individual characteristics rather than common characteristics and are items that must be performed based on the individual’s clinical condition. The FP-GROWTH algorithm is an effective method for finding frequent patterns in data sets. It is an improved algorithm based on the Apriori principle, but compared with the traditional Apriori algorithm, the FP-GROWTH algorithm is more efficient in processing large amounts of data.

Fourth, FP-GROWTH generates multiple paths, calculates the distribution probability of all data, converts patient data into a normal data distribution in different ‘diagnosis+therapy’ groups and realises data merging and clustering through Gaussian clustering. At the same time, the time dimension is superimposed, and the item is divided into three stages: before, during and after operation. The item with the highest number of frequent item combinations was selected and labelled as high, medium or low cost for the output result. Finally, a similarity calculation plays a calibration role in the model, and cosine similarity and Pearson correlation coefficients are used for different path data to study the conformity and fit degree between each path and patient data, to realise data calibration (figure 1).

Figure 1. Workflow of data processing.

Results

The conceptualisation and purpose of CEBPWs

CEBPWs are diagnosis and therapy standards for diseases established by concretising large amounts of real-world clinical data through big data technology. CEBPWs are built on DIP groups and take into account individual patient characteristics and reflect real-world data characteristics.

They have two main objectives. The first is to regulate medical practices, reduce costs and improve the quality of medical services; the second is to establish a correlation system of payment standards, diagnostic standards and therapy standards by forming a CEBPW application model incorporating comparable value, assessable quality, controllable process and standardised operation among payment standards, diagnostic standards and therapy standards. The CEBPWs also serve as a benchmarking system for hospital management and healthcare insurance payment and supervision.

Key elements of CEBPWs

CEBPWs have five elements. The first key element is the identification of objective evidence—the use of real-world data features to concretise practical problems that need to be solved in clinical practice and management. The objective of developing CEBPWs is to integrate the diagnostic standards and therapy standards of the disease group with the DIP payment standards and match the cost payment with the diagnostic and therapy content of the disease group, including key diagnostic and therapy items and their structure, the order of item classification, item frequency, cost weight and time and sequence. It proceeds by clustering, classifying and summarising clinical data to yield data evidence. A variety of data evidence is collected to form evidence clusters, and logically related evidence is integrated into big data clinical evidence-based pathways around problems to be solved.

The second key element is the correct classification of evidence. The objective and reasonable classification of large amounts of chaotic clinically produced data can convert it into evidence. Evidence can be divided into two categories: diagnosis and therapy. Diagnosis items can be subdivided into imaging, examination, endoscopy, pathology, nuclear medicine, etc, and therapy items can be divided into surgery, anaesthesia, intervention, drugs and consumables. An evidence bank was established according to the classification of diagnosis and therapy items, and the evidence was integrated into CEBPWs for different disease groups according to different needs.

The third key element is evidence value weight. Different ‘diagnosis+therapy’ groups have different diagnosis and therapy methods, which involve a large number of diagnosis and therapy items. As the diagnosis and therapy items pose huge challenges to medical insurance payment and medical management, it is necessary to analyse the cost of each item based on data. According to a few key principles, the diagnosis of a DIP group accounts for 80% of the total examination cost, corresponding to 20% of the examination items, and these items were clustered and summarised as the evidence of diagnosis value weight. The evidence of therapy value weight is also formed in the same way for therapy items. These key core items must exist in the corresponding disease group to standardise the disease diagnosis and therapy process, ensure the quality of care and objectively fit the cost and medical technology level. Using weight of evidence value makes the formulation of CEBPWs simplified, and the use of data is more targeted and applicable.

The fourth key element is evidence frequency weight. In a ‘diagnosis+therapy’ group, different examination items have different probabilities of appearing; those with a high probability of appearing in the cases of the DIP group will be used to form frequency evidence through probabilistic data feature mining. High-frequency items are essential to the diagnosis and therapy process of the ‘diagnosis+therapy’ group and are of great significance to the management of diagnosis and therapy norms and the payment and supervision of health insurance. If they are missing, the quality of diagnosis and therapy will be affected. Evidence frequency weight can be analysed to identify missing items or frequency anomalies in clinical practice.

The fifth key element is temporal ordering evidence. Clinical diagnosis and therapy are in a strict chronological order, with a progressive relationship between diagnosis items and therapy items on a timeline. At the same time, each day of hospitalisation corresponds to different diagnosis items and therapy items. According to the diagnosis and therapy process, the temporal data characteristics of diagnosis items and therapy items are mined to form temporal evidence. The diagnosis and therapy process of each ‘diagnosis+therapy’ group is divided into three stages: diagnostic examination, treatment intervention and recovery observation, and different temporal evidence pertains to each stage, thereby standardising diagnosis and therapy behaviours and increasing their accuracy (figure 2).

Figure 2. CEBPWs of cardiac surgery and superimposed time dimension. CEBPWs, clinical evidence-based pathways.

Results of CEBPWs

We shall take orthopaedic musculoskeletal system surgery as an example. Criteria for the CEBPWs for musculoskeletal system surgery include a clinical laboratory test, clinical immunological test, clinical chemistry test, clinical haematology test and colour Doppler ultrasonography, histopathological examination and X-ray inspection, artificial joints, general anaesthetic, opiates, peripherally acting muscle relaxants, hypnotics and sedatives and other beta-lactam antimicrobials. The completion of musculoskeletal system surgery should include the key items mentioned above (figure 3). Each column displays the items of medical treatment, including diagnosis, therapy, drugs and consumable items. Based on diagnosis, therapy, drugs and consumable items, standardised pathways are generated for mandatory and high-frequency items, while different CEBPW pathways are generated based on the patient’s individual characteristics and needs.

Figure 3. CEBPWs of orthopaedic musculoskeletal system surgery. CEBPWs, clinical evidence-based pathways.

Application result

We established an evaluation and supervision system based on the CEBPWs. Through comparison of the CEBPWs of the DIPs and practice, an early warning judgement was formed and abnormal therapy modes and cost structure deviations were evaluated to implement precise hierarchical supervision at the institutional, departmental, doctor, patient group and case levels, where the smaller the granularity, the higher the classification precision, facilitating accurate supervision.

Next, we established a mechanism for detecting suspected violations. By combining the macro-operation of the medical insurance fund with the specific use of the fund, we have gradually constructed an intelligent monitoring model for medical insurance based on big data. Specifically, with regard to issues such as up-coding, cost deviation, disaggregated hospitalisation and cost transfer, a mechanism for the detection of suspected irregularities has been established based on the CEBPWs.

CEBPWs were implemented and applied in Shanghai, and 42 954 pathways were formed. In 2023, 79 hospitals (24 tertiary hospitals and 55 secondary hospitals) in 16 districts of Shanghai conducted self-inspections through CEBPWs and the DIP Intelligent Supervisory System. Violations were found in 433 cases, including 71 cases of upcoding, 106 cases of cost transfer, 117 cases of overtreatment and 139 cases of other violations, involving a total cost of ¥6.95 million. The item frequency deviation was 8.64% and the cost deviation was 10.82%, with 8.95% for diagnosis, 9.44% for therapy, 14.81% for drugs and 8.98% for consumables; cost overruns were 6.02% and negative values were 4.80%; and the chronological deviation was 39.63%. The deviation between CEBPWs and CPWs is relatively small, it indicates that CEBPWs include most of the items in the clinical pathway, and, therefore, the pathway generated by CEBPWs is reasonable.

Discussion

In the era of big data, a good clinical pathway can make full use of digital technology and big data, specify diagnosis and therapy standards in the form of data characteristics and combine treatment data with payment standards to standardise clinical diagnosis and therapy behaviour by establishing a built-in mechanism, improve the accuracy and precision of diagnosis and therapy and strengthen medical quality control. With this in mind, we developed a novel clinical pathway, big data clinical evidence-based pathways. To our knowledge, we have pioneered CEBPWs worldwide, which can accurately reflect resource consumption and are transparent, objective and adaptable. Compared with CPWs and other studies, the clearest distinguishing feature of the CEBPWs lies in its use of data-driven techniques. CPWs have been modified and improved internationally to improve diagnostic excellence and resource stewardship; for example, researchers have suggested that future clinical pathway construction should pay more attention to diagnostic processes and feedback from clinician end-users,¹⁶ accelerate the development of digitally enhanced models of care,^{17 18} use modularity to resolve conflicts between a standardised system of CPWs and the provision of patient-centred heterogeneous care¹⁹ and incorporate national guidelines and recommendations into routine care practices.²⁰ Pugh-Bernard renovated the CPW programme using a structured method to improve the pathway development process, identify and address gaps, increase efficiency in the cycle time to build a pathway, boost multidisciplinary participation and integrate CPWs into the electronic health record system and increase pathway utilisation.²¹

Around the world, upcoding in Medicare has been a topic of interest to economists and policymakers. For example, studies in Germany,²² the USA²³ and Portugal²⁴ have shown that medical institutions obtain additional compensation by upcoding. Although there are numerous ways to monitor upcoding, these methods are statistical and do not relate to the overall healthcare regulatory system. Our CEBPWs can open the black box of upcoding. In previous studies of specialty care, it has been observed that 30% of inpatient antimicrobial therapy, 26% of advanced imaging and 12% of acute percutaneous coronary interventions are unnecessary or inappropriate. Reducing overtreatment has important implications for improving patient-centred care.25,30 Through CEBPWs, we can find cases of excessive diagnosis, therapy, drugs and consumable items, calculate the difference and deviation between the case cost and the standard cost of the DIP group and find the specific points of overdiagnosis and therapy.

Conclusion

Combining clinical big data and digital technology, we have developed a new clinical pathway called CEBPWs. This study demonstrates the efficacy of CEBPWs as an innovative hospital management system that integrates objective data-driven criteria with digital technology. CEBPWs address critical limitations of CPWs, such as over-reliance on expert experience, relatively slow and unable to display visualisations and the lack of integration between diagnostic/therapeutic standards and payment regulations. CEBPWs provide a robust mechanism for real-time monitoring, early warning and quality improvement in healthcare delivery. This can provide a reference for the future development of intelligent, rapid and objective CPWs, promoting the improvement of medical service quality, the rational use of medical insurance funds and the rational allocation of health resources.

Data availability statement

No data are available.