Understand your data's limitations. That advice comes from John Trabinger, IS manager at the Health Insurance Commission. And he should know after a successful application development project which is now delivering real-time reports over an intranet.
With 15 years' SAS experience under his organisation's belt, John Trabinger says it's no boast to say it has four or five of the top SAS people in Australia -- including two he'd happily pitch against anyone else in the country. But looking back on an extremely successful application development project aimed to deliver real-time reports over an intranet, Trabinger knows all the SAS skills and experience in the world would have counted for nothing without an intimate understanding of the data concerned. And he has nothing but disdain for the big bang approach to application development, claiming the project would have failed dismally if his team had tried to throw the entire corporate database at it from the start. The Health Insurance Commission's (HIC's) information section manager, Trabinger has just overseen development of the application for the HIC's Doctor Shopper project, first funded in the 1996/97 Federal Budget. Like any decent IT project, the development took longer than it should have, had many more teething problems than necessary, and is now inspiring HIC to apply the technology and application to other projects throughout the organisation. And Trabinger says the experience gained and lessons learned will serve the HIC well into the future.
An extremely complex social and medical issue, doctor shopping imposes a cost burden on the Medicare program and Pharmaceutical Benefits Scheme and, with prescription drug abuse and trafficking increasing around the world, can have an immense community cost. Under the HIC definition, doctor shoppers are people who attend more doctors and receive more medicine than necessary for their therapeutic needs. Many of the hard-core shoppers in Australia are drug-dependent, but a small criminal element obtains drugs to sell to others either here or in war-torn countries overseas. Limited resources mean that before the HIC takes action a doctor shopper will have seen 15 or more different general practitioners in a single year -- probably without any one doctor being aware they are seeing the others. They will also have had more than 30 Medicare consultations and obtained considerably more PBS prescriptions than clinically necessary. The HIC's figures show 12,240 doctor shoppers in 1995/96, falling to 9515 by 1997/98. Australia's most determined doctor shopper saw an incredible 620 different doctors in a single year, while as a group the most active 25 per cent see doctors on about 200 days a year, travelling widely and seeing many different practitioners on a single day. Interestingly, the fewest doctors write the most scripts for doctor shoppers, with just 5110 doctors writing 80 per cent of all doctor shopper scripts. Some 79 per cent of doctor shoppers shop for benzodiazepines, codeine combinations, narcotics, or benzodiazepines plus codeine combinations.
On the face of it, the simplest way to deal with doctor shoppers would be to alert every doctor they see that he or she is not the only love in their lives.
Privacy legislation makes that impossible. That's where the doctor shopper unit comes in. "Obviously doctors are quite intuitive when people have drug problems and can usually pick them up," Trabinger says. "But the idea is once we identify a patient we try and encourage the patient to be managed by one particular doctor." He says the HIC staff who deal with the doctor shoppers have the most unenviable job in the commission. The pharmaceutically-trained specialists work in doctor shopping teams all over Australia. "They are the ones who have to go around and knock on [doctor shoppers'] doors and try and encourage them to sign up to get their drug-dependence problem managed. [The HIC staff] are pretty special people," he says. When a doctor shopper is detected they are encouraged to sign a privacy release form and then settle on a single doctor to manage their dependency. The HIC specialist can then furnish that doctor with the patient's complete medical records. If more sinister criminal motives are believed to be behind their activities the specialists can also work with the compliance investigation unit, whose job is to liase with investigative bodies.
The drugs that doctor shoppers prefer can be dangerous. There will be times when lives will inevitably be at risk, and that makes getting accurate timely information to doctor shoppers and doctors absolutely essential. "In the old system the states would request a doctor shopper report; this would come through our normal report production team and be prioritised with all the other reports we do. You would write a batch job to pull the data out and the turnaround time would be four, maybe five days sometimes; and depending on priorities that particular week, it could be up to two weeks," Trabinger says.
Since it was obvious there was a need to deliver the information in real time, the HIC began a pilot project to try to produce the standard doctor shopping reports over an intranet. In successfully doing so, it has managed to cut turnaround time from five days to 30 seconds.
Long-time User
Predominantly a mainframe-based organisation, the HIC has only migrated to a PC client/server environment over the last few years. However, it has been a SAS user for at least 15. The doctor shopper project would involve finding ways to get information dynamically to operational staff. "The idea was to try to establish that we were going to be able to meet all those requirements in creating the many different hierarchies of the dimensions in the data. And a lot of that was obviously iterative, because we hadn't done it before and hadn't had that experience, but we had the SAS Institute here helping us do that," Trabinger says.
With no experience in intranet work, writing HTML, building a Web server, writing code for the Web server and developing multidimensional databases, the HIC did what it always did. It accepted it would need to help its own staff develop the skills necessary and worked hard to do so. That was the beauty of the doctor shopper project: while it is an extremely high profile project for the HIC to administer, the size of the database was small compared to the massive corporate database. This made it vastly easier to manage. But Trabinger says the team faced a much bigger challenge than simply giving its experienced staff new skills -- and that was getting the data right. "We put a lot of time and effort into planning how the data should be structured. To be honest, you can have all the fancy tools and you can have all the best ideas in the world, but unless you structure your data correctly you might as well forget it," he says. "In any information delivery project that is where most of the work gets put in -- structuring your data so it can be usefully analysed in the format you want it to be."The HIC is one of biggest number crunchers in the country, and a lot of that data sits on the mainframe. For the doctor shopper project the team built a doctor shopping database, then structured it carefully and put it on the intranet server. Trabinger says the database would continually grow into the future because of the need to create a different dimension for each different report requirement. "There are some limitations in this, in that we can't have a different dimension for every single report we want," he says. "We have to make sure we get the right ones out, and we will still have to go to the raw source and extract data for ad hoc-type requirements. But this way we are covering the main ones, the main specific-purpose requirements, the ones that will assist in operational day-to-day requirements." He says structuring the data was fundamental to the entire exercise. Without well-structured data all the tools available on the market would have been next to useless. Trabinger says the commission spent a lot of time working out the primary data variables and the levels at which these should be aggregated. After a first stage pilot that concentrated on reproducing current standard reports, a second stage began which involved providing more sophisticated dynamic reports with drill down OLAP capabilities. Geographical hierarchies providing drill down by state and geographic region have proved particularly useful in highlighting concentrations of doctor shoppers in particular areas.
Neural Networks
Now the HIC is looking for new ways to extend the value of the information. For several years it has been a world leader in the use of neural networks for fraud detection, using SAS to extract and manipulate data that is then fed into a proprietary neural network application. In fact, in the early 1990s, well before SAS and IBM were building neural networks, the HIC had recruited academic specialists in data mining to run neural network programs. It will now use neural network applications to detect patterns in unstructured dimensions of the doctor shopper application. "We haven't done this yet, but we think it will be very useful," Trabinger says. "We believe that multidimensional databases will assist the neural net because the data will be structured in more ways than we have used the neural net to analyse the data. That is something that we are quite excited about, and we're keen to explore that avenue of having different hierarchies and dimensions of data and running that through the neural net.
"Basically you use neural nets to find patterns that you don't know exist. When you have a look at the data some of the patterns are obvious, but you can use a neural network to train, to learn those patterns, so when you throw data to it the next time, it will automatically show you those patterns and the discrepancies," he says. But he says until the HIC understands these new data patterns thoroughly and has a thorough understanding of the desired outcomes, it will be impossible to achieve some of the desired projects. An information-driven division and organisation where people need vast amounts of information at the operational level, the HIC sees a lot of application for the technology -- not only in getting information out to people in real time, but also in being able to manage their activities and projects remotely. The HIC also sees the technology as the key to empowering those charged with analysing data, helping health knowledge workers to make faster and better decisions. "We have to be smart about that. We have to go back to the drawing board and ask how we can create structured data that will let these knowledge workers and decision-makers do the analytical work that they need to do. That is something that will be exciting over the next 12 months," Trabinger says. But first his team will again work to achieve an intimate understanding of any data they work with. "You are nothing without the data and knowledge of the data," he says.
"When people tell me their organisation has got great data, I know that they don't know their data well at all, because not one database in this country isn't flawed with some inadequacy."Likening the relationship with the corporate data to a marriage, Trabinger says an organisation wanting to work with its data to achieve useful results not only had to know its data intimately but also know its limitations. "Every database has limitations. If you don't know those, you can't provide successful information," he says. "If anyone tells you they have got wonderful data, don't ever work on a project with them. If a CIO or IT manager says to me: 'Let's work on a project together, our data is really good', I would run a mile."Getting Sassy at the HIC Years of experience has taught John Trabinger, the Health Insurance Commission's (HIC's) information section manager, that it is useless to try to recruit senior SAS-experienced people on public service salary structure.
Therefore the HIC is happy to develop staff itself. It has also learned that within two to three years of taking on IT graduates to work with SAS, those people become the most marketable in the organisation. While that makes keeping SAS staff an ongoing challenge, Trabinger is reasonably satisfied with the program he has developed to deal with the problem. Trabinger has 28 SAS people working directly for him, including four who find the working conditions sufficiently satisfactory and the work challenging enough to make them happy to stay long term. "We are pretty lucky that we have those four or five people who can handle very complex problems using SAS applications. Without them we would be in trouble. You won't get anything done without good people," Trabinger says.
Keeping the other staff is made a little easier because so much of the work the HIC can offer is challenging and satisfying. "That excites people -- the work.
And the HIC, like other organisations, is looking at new award agreements. The public service is moving down a track where it will be able to reward people for their skills and experience, but at the moment the HIC is only investigating that possibility," he says. "Our people are very marketable. Even within the organisation, anyone who accepts someone from our section in the organisation knows they are getting someone with good technical skills and good business knowledge, because they are at the edge of the data the whole time." Every six months Trabinger goes through a significant recruitment drive. The HIC program makes it possible to bring graduates in at a low level and broaden their skills so that as soon as they reach a certain level of competency they can move on to the next level without a formal procedure. The program encourages them to achieve and to be prepared to learn in a learning environment. "This way they will probably stay a bit longer. If we put them on the higher level straight away, they would want the next level in a very short time. That is not fair to us and it is not fair to them, because they wouldn't be ready for that level and what we demand of people of that level," Trabinger says. "This way they get to come and get a reasonable level of core knowledge and experience before they move up, they have a career path and we can retain them." He is also very careful about who he recruits. Recruit people with IT skills only and they can lose interest very quickly and want to go on to more challenging things. Recruit people with a purely statistical background and you'll find you can offer them only so much stats work before they will be looking for more applied stats. And if you bring people in with business knowledge without the aptitude to learn the IT side of the equation, it presents its own challenges. "That means there is not one right-skilled person to do this sort of work. I look for people with a diverse range of skills and background. I bring in people with IT, Business and Stats degrees, and I find those three groups are rubbing off well against each other all the time," Trabinger says. "So there is not one right person to do this work; you really need a broad group. That has been a hard process of trial and error for us. You lose good people, but you have to take some enjoyment that you have developed people and given them the opportunity to develop themselves."
Join the CIO Australia group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.