To the Editor: The editorial by Coiera and Westbrook1 and indeed the letter by Fox2 tended to use the term “clinical software” in a broad sense.

In Australia, doctors who use clinical applications are in fact using electronic medical records. The major functionality provided is one of information storage, with the ability to produce a range of documents that were previously handwritten. To accept that the currently available applications offer decision support is a very generous, and possibly naïve, interpretation.

The common example of decision-support tools used in Australia is the humble prescription writer. Current vendors offer a variety of prescription writers and, as Coiera and Westbrook1 assert, they check for drug–drug interactions and dosage errors and provide various alerts.

Coeira and Westbrook go on to question whether appropriate testing is being performed on the large number of applications available. At first glance this question may seem to be somewhat invalid, as most of the software packages in Australia use either the AZDex (a proprietary internal drug database used by Medical Director) or MIMS (a pharmaceutical database of products currently available in Australia by CMPMedica Australia) drug databases. These two highly regarded sources of drug information provide the developer with an easy-to-implement set of tools that effectively ensures “quality” information is provided to the doctor preparing the prescription.

The problem is that, although we have quality databases, there is little or no compliance testing to ensure that the applications that use them are developed to an equally high standard. For example, there is no mechanism to inform end-users which parts of the database have been used, and there is no testing to ensure the end-user is presented with accurate information.

While many Australian doctors have moved to computerised clinical records, their ability to use these data for improving clinical care is being curtailed by a lack of standards and coding of conditions. Computers are not efficient in dealing with the free text that is traditionally used in clinical notes, and even data such as drug prescriptions are difficult to analyse because of the lack of a standard method of drug naming or coding.

I look to a future when true clinical support tools are available. To this end, the development, coordination, and facilitation of a series of standards by the National E-Health Transition Authority should be supported.

In reply: At the heart of much debate on patient consent for access to electronic data are two conflicting desires — many consumers wish to minimise access to their record, and many clinicians have genuine concerns that such restriction may lead to patient harm. In some cases, privacy is paramount (eg, psychiatric or sexual health history). In others, such as emergency presentations, patient wellbeing may override such concerns. This has led many to conclude that there is no “one size fits all” model for e-consent.1

The current debate between the boundary cases of “opt-in” and “opt-out” is misleading because many specialist services of necessity will have local consent processes, crafted to meet the need of their patients and their clinicians. Yet, many health information technology initiatives do not seem prepared to consider this complexity, and opt-in or opt-out are all that is on offer. Liaw and Boyle’s concerns about dropout rates under an opt-in system affecting secondary use of patient data for research purposes are no doubt real, but it is hard to draw too strong a comparison between patient recruitment for research and patient permission to store data for their own care.

Williams correctly points out in his letter that decision support remains a small component of the software to support clinical practice that most Australian general practitioners now use. However, anyone using a prescription program that suggests doses, checks interactions, or generates alerts is using decision support. We can say so confidently because research repeatedly shows that such functions change clinical decisions. Indeed, something as simple as accessing research articles and guidelines using the Internet is a form of decision support, because it changes clinical decisions significantly, and sometimes negatively.2 Consequently, it is perhaps naïve to await “true” decision support using artificial intelligence before we worry about how software affects clinical behaviour. If the intervention was a drug and serious patient harm resulted from infrequent side effects, everyone would quickly agree some controls might be needed. Somehow, we still don’t seem to get as excited about the harm that may come from using bread-and-butter clinical software, but we should.