Detecting hospital behaviors of up-coding on DRGs using Rasch model of continuous variables and online cloud computing in Taiwan

Chien et al. BMC Health Services Research https://doi.org/10.1186/s12913-019-4417-2 (2019) 19:630 RESEARCH ARTICLE Open Access Detecting hospital behaviors of up-coding on DRGs using Rasch model of continuous variables and online cloud computing in Taiwan Tsair-Wei Chien1, Yi-Lien Lee2,3 and Hsien-Yi Wang4,5* Abstract Background: This work aims to apply data-detection algorithms to predict the possible deductions of reimbursement from Taiwan’s Bureau of National Health Insurance (BNHI), and to design an online dashboard to send alerts and reminders to physicians after completing their patient discharge summaries. Methods: Reimbursement data for discharged patients were extracted from a Taiwan medical center in 2016. Using the Rasch model of continuous variables, we applied standardized residual analyses to 20 sets of norm-referenced diagnosis-related group (DRGs), each with 300 cases, and compared these to 194 cases with deducted records from the BNHI. We then examine whether the results of prediction using the Rasch model have a high probability associated with the deducted cases. Furthermore, an online dashboard was designed for use in the online monitoring of possible deductions on fee items in medical settings. Results: The results show that 1) the effects deducted by the NHRI can be predicted with an accuracy rate of 0.82 using the standardized residual approach of the Rasch model; 2) the accuracies for drug, medical material and examination fees are not associated among different years, and all of those areas under the ROC curve (AUC) are significantly greater than the randomized probability of 0.50; and 3) the online dashboard showing the possible deductions on fee items can be used by hospitals in the future. Conclusion: The DRG-based comparisons in the possible deductions on medical fees, along with the algorithm based on Rasch modeling, can be a complementary tool in upgrading the efficiency and accuracy in processing medical fee applications in the discernable future. Keywords: Rasch analysis, Standardized residual analysis, Dashboard, Medical center Background Fee-for-service (FFS) is a payment system, in which the health care providers are paid for each service performed [1]. To reduce the rapid growth rate of health expenditures, diagnosis-related groups (DRGs) are launched according to patients with similar clinical characteristics, resource consumption patterns, and comparable costs [2]. * Correspondence: 4 Department of Sport Management, College of Leisure and Recreation Management, Chia Nan University of Pharmacy and Science, Tainan, Taiwan 5 NephrologyDepartment, Chi Mei Medical Center, 901 Chung Hwa Road, Yung Kung Dist., Tainan 710, Taiwan Full list of author information is available at the end of the article Taiwan’s National Health Insurance (NHI) scheme, launched in 1995, originally used FFS. Despite various legislative and administrative measures aimed at capping maximum reimbursements, including a global budget system and a case-payment scheme, the rapid increase of medical expenses continued to occur [3, 4]. In response, the Bureau of National Health Insurance (BNHI) began using a Taiwan-specific DRG system (TW-DRG) in January 2010, and a total of 1663 TW-DRGs were developed until 2016. The main problem here is how to detect hospitals’ behaviors of up-coding on DRGs [5]. Traditionally, the BNHI adopts the peer-review approach by giving the © The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Chien et al. BMC Health Services Research (2019) 19:630 whole list of medical expenditures (or say items) to physicians from other hospitals for examining whether the reimbursement case is rational and reasonable. An efficient and effective detection method can be thus expected to improve using item response theory (IRT) modeling, particularly the one-parameter Rasch model [6] of continuous items [7, 8]. Because that (1) such cases of inpatient expenditures are continuous variables, (2) IRT-bases Rasch model is one-parameter simple model requiring relatively small sample size to calibrate model parameters, and (3) the Rasch model of continuous items [7, 8] has been developed before in the literature, an online routine (or say application programming interface, API) can be used for detecting abnormal upcoding behaviors on DRGs [5]. Importantly, the detection should be (and must be) objective and scientifical as much as possible. Given that the DRGs are characterized by similar resource consumption patterns and comparable costs [2], any up-coding in a discharge case results in items being miss-fitted to the model if the standardized residual analysis [9] is applied to inspect. That is, the Z-score on the response interacted by the momentum of the case and the item equals ðobserved−SDexpectedÞ, where SD = standard deviation on the item and the case [10, 11]. To solve the problem, two approaches are implemented in the current study: (1) verifying the effectiveness of the Rasch standardized residual analysis in terms of DRG detection, and (2) developing an online detection scheme for tracking any item misfitting to the model. The latter approach can help alert physicians once they accomplish their inpatient’s discharge summaries, allowing them to prepare the necessary notes (or actions) before the BNHI assessment on the reimbursement of medical expenditures has been implemented. In this work, we aim to apply the Rasch model of continuous variables [7, 8] (1) to verify the effectiveness of detection on TW-DRGs, and (2) to develop an online checking tool for selecting the most misfit items on TW-DRGs for each inpatient case. Methods Data source Experimental and control groups We applied the TW-DRG classification module issued by the BNHI to two groups, namely, the control and experimental groups. Control group A set of 300 cases (as norm-reference) from 20 TW-DRGs(i.e., like types of tests) were randomly selected from a medical center in southern Taiwan between 2015 and 2016 and were not deducted yet by the BNHI assessment for the medical fees on any Page 2 of 7 item. These 300 cases were used for calibrating item (or say fee) parameters(i.e., item difficulties on IRT terms) as references comparable to the experimental group. Experimental group We randomly selected 194 cases on the 20 TW-DRGs mentioned above from the studied medical center at the same period(i.e., 2015 and 2016). Data on these 194 c (...truncated)