Data Mining in Healthcare

Current Applications, Issues and Fraud detection

By: Ritu Tayal Humana 249936

Abstract
The successful application of data mining in highly visible fields like e-business, marketing and retail have led to the popularity of its use in knowledge discovery in databases (KDD) in other industries and sectors. Among these sectors that are just discovering data mining are the fields of medicine and public health. Here, we would be studying the current techniques of KDD, using data mining tools for healthcare and public health. We would also discuss critical issues and challenges associated with data mining and healthcare in general. There are a growing number of data mining applications, that are capable of analysis of health care centers for better health policy-making, detection of disease outbreaks and preventable hospital deaths, and detection of fraudulent insurance claims.

1) Introduction
The successful application of data mining in highly visible fields like e-business, marketing and retail have led to its application in KDD in other industries and sectors. Among these sectors just discovering IT healthcare. A study on the current techniques of knowledge discovery in databases using data mining techniques that are in use today in medical research and public health has been given here. Moreover, a discussion on some critical issues and challenges associated with the application of data mining in the profession of health and the medical practise in general has also been made. 1.1) Objectives

To figure out the current uses and highlight the importance of data mining in medicine and public health. To find data mining techniques used in other fields that may also be applied in the health sector. To identify issues and challenges in data mining as applied to the medical practise. To outline some recommendations for discovering knowledge in electronic databases through data mining.

2) Research Findings
2.1) Data Mining in the health sector The practise of using concrete data and evidence to support medical decisions (also known as evidence-based medicine or EBM) has existed for centuries. John Snow, considered to be the father of modern epidemiology, used maps with early forms of bar graphs in 1854 to discover the source of cholera and prove that it was transmitted through the water supply. Florence Nightingale invented polar-area diagrams in 1855 to show that many army deaths could be traced to unsanitary clinical practises and were therefore preventable. Snow and Nightingale were able to personally collect, sift through and analyze the

mortality data during their times because the volume of information was manageable. Today, the size of the population, the amount of electronic data gathered, along with globalization and the speed of disease outbreaks make it almost impossible to accomplish what the pioneers did. This is where data mining becomes useful to healthcare. It has been slowly but increasingly applied to tackle various problems of knowledge discovery in the health sector. Data mining and its application to medicine and public health is a relatively young field of study. Authors refer to data mining as the process of acquiring information, whereas others refer to data mining as utilization of statistical techniques within the knowledge discovery process. The generally accepted definition for Data Mining is Extraction of interesting knowledge based on certain rules, regulations, patterns and constraints of data from large databases. 2.2) Importance and Uses of Data Mining in Medicine and Public Health

Despite the differences and clashes in approaches, the health sector has more need for data mining today. Several arguments have been passed that support the use of data mining in the health sector, covering not just concerns of public health but also the private health sector. Data overload. There is a wealth of knowledge to be gained from computerized health records, but the data is so huge that it seems that one is being drowned inside the data and is actually starving for knowledge.In fact, some experts believe that medical breakthroughs have slowed down, attributing this to the prohibitive scale and complexity of present-day medical information. Computers and data mining are best-suited for this purpose. (Shillabeer and Roddick 2007). Evidence-based medicine and prevention of hospital errors. Medical institutions applying data mining on their existing data, discover new, useful and potentially lifesaving knowledge that otherwise would have remained lifeless in their databases. For instance, an ongoing study found out that 87% of hospital deaths in the United States could have been prevented if precaution would have been taken in commiting errors. By mining hospital records, such safety issues could be flagged and addressed by hospital management and government regulators. Policy-making in public health. GIS and data mining combined tools like (free, open source, Java-based data mining tools), to analyze similarities between community health centers in Slovenia. After mining the data, certain patterns were discovered among health centers that led to policy recommendations to their Institute of Public Health. Therefore, it was concluded that data mining and decision support methods , including novel visualization methods, can lead to better performance in decisionmaking. More value for money and cost savings. Data mining allows organizations to get more out of existing data at nominal extra cost. KDD and data mining have been applied to discover fraud in credit cards and insurance claims. By extension, these data mining techniques could also be used to detect anomalous patterns in health insurance claims.

Early detection and/or prevention of diseases. Use of classification algorithms has been cited to help in the early detection of heart disease, a major public health concern all over the world. Data mining tools provide aid in monitoring trends in the clinical trials of cancer vaccines. By using data mining and visualization, medical experts could find patterns and anomalies better than just looking at a set of tabulated data. Early detection and management of pandemic diseases and public health policy formulation. Health experts have also begun to look at how to apply data mining for early detection and management of pandemics. This could be achieved by combining spatial modeling, simulation and spatial data mining to find interesting characteristics of disease outbreak. The analysis that resulted from data mining in the simulated environment could then be used towards more informed policy-making to detect and manage disease outbreaks. Non-invasive diagnosis and decision support. Some diagnostic and laboratory procedures are meddlesome, costly and painful to patients. An example of this is conducting a biopsy in women to detect cervical cancer. Therefore, K-means clustering algorithm to analyze cervical cancer patients came into picture and found that clustering gives better predictive results than existing medical opinion. A set of interesting attributes could be used by doctors as an additional support on whether or not to recommend a biopsy for a patient suspected of having the cervical cancer. Adverse drug events (ADEs). Some drugs and chemicals that have been approved as nonharmful to humans are later discovered to have harmful effects after long-term public use. Finding realities of this, US Food and Drug Administration uses data mining to discover knowledge about drug side effects in their database. This algorithm called MGPS or Multi-item Gamma Poisson Shrinker was able to successfully find 67% of ADEs five years before they were detected using traditional ways. This is how data mining applications are used in early detection of diseases, prevention of deaths, improvement of diagnoses and even detecting fraudulent health claims. However, there are warnings to the use of data mining in healthcare.

3) Issues and Challenges

Applying data mining in the medical field is quiet a challenging engagement due to the peculiarity of the medical profession and several inherent conflicts between the traditional methodologies of data mining approaches and medicine. In medical research, data mining starts with a hypothesis and then the results are adjusted to fit the hypothesis. Such kind of proceedings diverges from standard data mining practise, which simply starts with the data set without an apparent hypothesis. Also, whereas traditional data mining is concerned about patterns and trends in data sets, data mining in medicine is more interested in the minority that do not conform to the patterns and trends. This difference in approach gets magnified as most standard data mining is concerned only with description and not eloboration of the patterns and trends. In contrast, medicine needs those explanations because a slight difference could change the balance between life or death. For example, anthrax and influenza share the same symptoms of respiratory problems. Lowering the threshold signal in a data mining experiment may either raise an anthrax alarm when there is only a flu outbreak. The converse is even more fatal: a perceived flu outbreak turns out to be an anthrax epidemic. Even if data mining results are credible, convincing the health practitioners to change their routine habits based on evidence may be a bigger problem. Doctors in hospitals refused to change hospital policy even when confronted with evidence. Even being presented with evidence, doctors still refused to change their habits until only much later. Privacy of records and ethical use of patient information is also one big obstacle for data mining in healthcare. For data mining to be more accurate, it needs a sizeable amount of real records. Healthcare records are private information and yet, using these private records may help stop fatal diseases.

4) Fraud Detection
4.1) Healthcare Fraud detection

Fraudulent healthcare claims increase the burden to society. Therefore, healthcare fraud detection becomes important. Generally, healthcare frauds are not obvious and thus difficult to detect. The followings are typical examples of healthcare fraud techniques used by health care providers and patients: Providers billing for services not provided. Providers administering (more) tests and treatments or providing equipments that are not medically necessary. Providers administering more expensive tests and equipments (up-coding). Providers multiple-billing for services rendered. Providers unbundling or billing separately for laboratory tests performed together to get higher reimbursements. Providers charging more than peers for the same services. Providers conducting medically unrelated procedures and services. Policy holders traveling long distance for treatment which may be available nearby. Policy holders letting others use their healthcare cards.

4.2)

Statistical healthcare fraud detection techniques

The net effect of excessive fraudulent claims is excessive billing amounts, higher perpatient costs, excessive per-doctor patients, higher per-patient tests, and so on. This excess can be identified using special analytical tools. Provider statistics include: 4.3) Total amount billed. Total number of patients. Total number of patient visits. Per-patient average billing amounts. Per-patient average visit numbers. Per-patient average medical tests. Per-patient average medical test costs. Per-patient average prescription ratios (of specially monitored drugs). Analytic healthcare Fraud detection methods

Healthcare fraud detection involves account auditing and detective investigation. Careful account auditing can reveal suspicious providers and policy holders. Ideally, it is best to audit all claims one-by-one carefully. However, auditing all claims is not feasible by any practical means. Furthermore, it's very difficult to audit providers without concrete smoking clues. A practical approach is to develop short lists for scrutiny and perform auditing on providers and patients in the short lists. Various analytic techniques can be employed in developing audit short lists. Keep in mind that excessive fraudulent claims lead deviations in aggregate claims statistics. In addition, fraudulent claims often develop into patterns that can be detected using predictive models! 4.4) Statistical listings of risky providers

When unkind claims are repeated frequently, it leads to higher provider statistics. Various provider statistics can be used to identify fraudulent claims. For instance, audit short-lists may include the followings: Doctors who treated whopping, say, 50+ patients in a day. Providers administering far higher rates of tests than others. Providers costing far more, per patient basis, than others. Providers with high ratio of distance patients. Providers prescribing certain drugs at higher rate than others.

It is noted that statistical analytic techniques can reveal excessive providers who might be outright stupid! But it will be difficult to identify modest level fraud activities. The subsequent section describes how sophisticated techniques can be applied. 4.5) Data Mart for healthcare claims audit

Incorporating the techniques described in previous sections leads to intelligent audit and fraud detection environment. It is noted that healthcare fraud detection requires compilation of potentially huge data, involving complex computation and sorting operations. Our data mart platform for healthcare fraud detection is based on the following architecture.

First, claim payment records are transformed and loaded into healthcare fraud data mart. Data is added into data mart, normally monthly or quarterly basis. Summary information is created for providers, doctors and policy holders. Expert systems engines are used to analyze, score and detect potentially risky providers and claims, Finally, auditors (and investigators) analyze data.

Data Transformation Loading

Data Mart

BI Analytics

Claims records

Predictive Rules

5) Conclusions and Recommendations

This study on data mining applications in medicine and public health provides an overview of current practises and challenges healthcare is facing.These applications can help the healthcare organizations and agencies in extracting knowlegde from their datawarehouses available with them. Before embarking on data mining, however, an organization must formulate clear policies on the privacy and security of patient records. It must enforce this policy with its partner stakeholders and its branches as well as agencies. Public health concerns like rapid pandemic outbreaks, the need to detect the onset of disease in a non-invasive, painless way, and the need to be more responsive to its customers all such concerns adds up to an increasing need for health organizations to integrate data and apply data mining to analyze these data sets.