AI in Health Financing

--- title: 'Project documentation template' disqus: hackmd --- Project Title === ![downloads](https://img.shields.io/github/downloads/atom/atom/total.svg) ![build](https://img.shields.io/appveyor/ci/:user/:repo.svg) ![chat](https://img.shields.io/discord/:serverId.svg) ## Table of Contents [TOC] Background --- Doing research on AI implementation in health financing at Taiwan Goals --- Implementation of AI in: 1. Review BPJS Claim data based on patients' Clinical data  Literature Review --- **Taiwan's National Health Insurance (NHI) use AI technology as a tool to review claim data** (src: https://www.mohw.gov.tw/cp-4254-49119-1.html created in 2019/08/29) Taiwan's National Health Insurance (NHI) system has indeed made significant progress in developing Natural Language Processing (NLP) models for the automated interpretation of radiology diagnostic report, such as CT scans. This application of AI aims to improve the accuracy and efficiency of the claim review process and promote better healthcare outcomes. Since the establishment of the NHI, it has included numerical structured data on national claim data of medical expenses, of which 360.000.000 claim for outpatient case and 3.440.000 for inpatient case are reported each year. Since 2014, some specific healthcare institutions have been encouraged to upload lab exam report and other unstructured text-based report, until July 2019, have accumulated a total of 2.470.000.000 lab reports, which is very large amount of data. In order to improve the accuracy and efficiency of the review mechanism in health insurance medical expenses, including to reduce the burden of 3,500 reviewers who need to complete about 2.6 million professional review cases each year, the medical exam report data can helps in evaluating the necessity of the medical exam perfomed on each patient more comprehensively, also accurately screening of hospitals or physicians with abnormal execution (dentifying potentially problematic claims for further investigation) before going through professional reviewer in validate the claims reasonably, which is increase the review efficiency. However, a large amount of medical report data requires the use of AI technology in order to organize it into usable information. Based on NHI analysis of big data, the item with the highest medical expenses in various medical examination performed in patient is CT scan, and head is the most frequently performed for the inspection site (approximately 40% of total CT scan performed). Therefore, NHI prioritize the implementation of AI for the analysis of radiology report, particularly focusing on head CT scans in developing the NLP model, to train machine learning experts to mark annotation and interpret the results, using the 1,000 selected CT head radiology report data uploaded by the hospital in the second quarter of 2017 as data model, train the AI to perform the tasks of lesion/ annotation marking, labeling, and report classification. The results show that the accuracy rate of the analysis results of the NLP machine learning analysis model is 99% compared with the results of expert interpretation. Using the NLP model to analyze 140,000 head CT scan reports (for data in fourth quarter of 2017), the results showed that about 40% of the radiology exam results are normal (positive result without any disease detected), it may be caused by disease exclusion or after treatment follow-up result (where patient may have recover from the diseased), but may also caused by unnecessary radioloy screening. According to the analysis of big data in health insurance, specifically related to CT and MRI exams performed in outpatient clinics of designated hospitals in Taiwan, reveals interesting findings. In 2019 (January-June), approximately 23,000 cases (3% of the total cases), were identified with abnormal radiology reports. These cases were associated with primary care diagnoses such as headaches, arthritis, cough, and other similar conditions. To address this issue, Taiwan's National Health Insurance (NHI) system takes proactive measures by reporting these abnormal radiology reports to the physicians in the respective hospitals. This collaboration with the medical community aims to reduce unnecessary and wasteful examinations. Additionally, the NHI system utilizes a self-developed NLP model for auxiliary analysis, enhancing the capabilities for further investigation. The National Health Insurance (NHI) in Taiwan recognizes the importance and potential of developing AI intelligence as a review tool for medical claim data. The use of AI, particularly through NLP models, has demonstrated significant advantages in terms of efficiency and time savings compared to real-world expert interpretation. The NLP model developed by the NHI has been able to analyze 140,000 data in just 10 hours, averaging around 0.25 seconds per data point. In contrast, it would take approximately 13 months for real-world experts to interpret the same amount of data, averaging around 4 minutes per data point. The NHI's estimation that 960 experts would be required to interpret data at the same speed as the NLP model emphasizes the need for AI tools in the review process. The NHI's plans to expand the development of AI intelligent review tools to include chest and abdomen radiology reports indicate a forward-looking approach. National Health Insurance (NHI) in Taiwan is committed to continuing the application of big data and AI technology to improve healthcare services. By leveraging these technologies, the NHI aims to develop a smart medical service review system that benefits both patients and doctors, while respecting the expertise of the medical profession. **Erosion of health insurance funds, excessive medical treatment in China** (src: https://en.wikipedia.org/wiki/Healthcare_in_China) In 2016, a study reported that a large number of doctors and patients conspired to erode medical insurance funds in China.[49] Several media have disclosed that the means of eroding the medical insurance fund include farmers being “hospitalized”, treating patients without illness, falsely increasing the number of days a patient is hospitalized, fake medication, fake surgery, excessive examinations, serious treatment of minor illnesses, repeated charges, and failure to provide Charges for services (empty charges), listing surgical treatment expenses that are not within the scope of reimbursement as reimbursement, retail pharmacies swiping medical insurance cards to sell daily necessities.[49][46] However, without the use of medical insurance funds, some medical institutions will be unsustainable, and may not be able to pay wages or repay loans. Affected by all parties, the regulatory authorities often cast their doubts on the rat.[50] In the whole year of 2019, China's medical insurance departments at all levels inspected a total of 815,000 designated medical institutions, and investigated and dealt with 264,000 medical institutions that violated laws and regulations; a total of 33,100 people who participated in violations of laws and regulations were dealt with, and a total of 11.556 billion yuan was recovered.[51] At the same time, over-diagnosis, over-examination, and over-medication in the medical industry have become common phenomena due to the loss of profitability of hospitals and the supply of medicines, which wastes medical resources.[49] In response to these problems, various localities have begun to coordinate and supervise medical insurance funds;[51] explore the introduction of medical insurance intelligent monitoring systems to intelligently review medical insurance funds; centrally purchase pharmaceutical consumables to save money and increase the proportion of medical labor technology value in fund settlement. *AI empowers medical insurance in China using the concept of DRG and DIP reimburse approach* (https://cnews.chinadaily.com.cn/a/202209/28/WS633400dea310817f312f0430.html created in 2022/09/28) Discussion --- <table> <tr><td> Discussion date: 2023/07/06 - Most of implementation of AI in Taiwan focus on clinical application in order to assist with clinical decision-making. For example, implementation of lesion detection and annotation marking in radiology image. - In the field of AI in health financing, China in the progress of implementing AI in their national insurance system, called E-bao - In the scenario of Mammography screening for national insurance claim in Taiwan, the insurance fee can be claim as long as the radiology report available. Some country will do double check for the insurance claim scenario. - HIS in Taiwan developed by each internal hospital - Institute of Brain Science in National Yang Ming Chiao Tung University project highlight: Brain tumor segmentation and classification, Coronary Artery Calcium (CAC), detection of Tuberculosis - Next sharing session on 2023/07/20 (Thursday) 19:00-21:00 Indonesia Time Topic: Implementation of AI in Brain Tumor Detection (Lesion delineation framework for vestibular schwannoma, meningioma, and brain metastasis for using stereotactic MR images) Speaker: Yu-Te Wu (Professor Institute of Biophotonics, National Yang Ming Chiao Tung University (NYCU), Taiwan) Outlines: • Automatic Segmentation of vestibular schwannoma, meningioma, and brain metastasis using deep-learning neural network • Automatically delineate tumor contour and estimate 3D volume after Gamma knife (GK) treatment • Centralized learning (CL) versus Federated learning (FL) • Detection of vestibular schwannoma • Outcome Prediction of GK Radiosurgery Using Radiomics • Detection and Segmentation for the Arteriovenous Malformation Lesions Using YOLOv5 and U-Net++ Published Works: * Journal of Therapeutic Radiology and Oncology: Applying Artificial Intelligence to Longitudinal Imaging Analysis of Vestibular Schwannoma Following Radiosurgery * Brain tumor AI Product Name: DeepBT Detector+ (using YOLOv5 for detection and U-Net++ for Segmentation) * Journal of Medical and Biological Engineering: Detection of Vestibular Schwannoma on Triple-parametric MRI Using Convolutional Neural Networks * Radiotherapy and Oncology: Prediction of pseudoprogression and long-term outcome of vestibular schwannoma after Gamma Knife radiosurgery based on preradiosurgical MR radiomics * 2022 TFDA win awards in Future Tech Award: Detection and Segmentation for the Arteriovenous Malformation Lesions Using YOLOv5 and U-Net++ * Collaborate with EBM, Taipei Veteran General Hospital (National hospital in Taiwan with the largr number of brain tumour patient treat in this hospital), References * [DeepBT demo](https://youtu.be/C_s7qNBO7yQ) * [China health insurance condition](https://cn.nytimes.com/business/20230224/china-health-insurance-explained/dual/) * [China health insurace type](https://www.chinagp.net/CN/10.12114/j.issn.1007-9572.2018.00.072) * [China Healthcare Security (CHS) manage country health insurance system] (https://en.wikipedia.org/wiki/Healthcare_in_China) </td></tr> </table> Methodology --- 1. Convert radiology free-type report into structured report ###### tags: `Templates` `Documentation`