Network Analysis and Detection of Health Care Fraud
You, Jong-Sung
In my earlier entries on “Statistics and Detection of Corruption��? and “Missing Women and Sex-Selective Abortion,��? I demonstrated that examination of statistical anomaly can be a useful tool for detection of crime and corruption. In these cases, binomial probability distribution was a very useful tool.
Professor Malcolm Sparrow at the Kennedy School of Government shows how network analysis can be used to detect health care fraud in his book, License to Steal: How Fraud Bleeds America's Health Care System (2000). He gives an example of network analysis performed within Blue Cross/Blue Shield of Florida in 1993.
An analyst explored the network of patient-provider relationships with twenty-one months of Medicare data, treating a patient as linked to a provider if the patient had received services during the twenty-one-month period. The resulting patient-provider network had 188,403 links within it. The analyst then looked for unnaturally dense cliques within that structure. He found a massive one. “At its densest core, the cluster consisted of a specific set of 122 providers, linked to a specific set of 181 beneficiaries. The (symmetric) density criteria between these sets were as follows:
A. Any one of these 122 providers was linked with (i.e., had billed for services for) a minimum of 47 of these 181 patients.
B. Any one of these 181 patients was linked with (i.e., had been “serviced��? by) a minimum of 47, and an average of about 80, of these providers.��?
After the analyst found this unnaturally dense clique, field investigations confirmed a variety of illegal practices. “Some providers were indeed using the lists of patients for billing purposes without seeing the patients. Other patients were being paid cash to ride a bus from clinic to clinic and receive unnecessary tests, all of which were then billed to Medicare.��?
Professor Sparrow suggests that many ideas and concepts from network analysis can be useful in developing fraud-detection tools, in particular for monitoring organized and collusive multiparty frauds and conspiracies.
Posted by Jong-sung You at 2:36 AM