Using AI to detect health fraud?

Will artificial intelligence algorithms help to inform clinical practice guidelines and fight against the possibility of fraud and over-servicing?

A recent article highlighted the use of software tools to detect fraudulent Medicare claims by a general medical practitioner. The case revolves around a general practitioner who had been altering patient records to cover up suspect billing practices. Late last year the NSW Health Care Complaints Commission found the doctor guilty of unsatisfactory professional conduct and professional misconduct, and cancelled his registration as a Medical Practitioner with a 3 year non-review period.

Publication of the case highlighted that the Professional Services Review (the agency that investigates Medicare-referred cases of possible inappropriate practice relating to Medicare, the CDBS and the PBS) had analysed more than 100 patient records using a piece of software called Parser. This custom software was designed to identify date discrepancies between the date of service and the date a related record was created. In this case it found that records were altered significantly after the date of the actual patient consultations. Concerns had been raised that artificial intelligence software was being used in compliance checks.

Apparently PSR has been using this software technology for around five years – a fact that only came to light because the HCCC case specifically referenced Parser in their findings. According to reporting on the issue, this specific software tool does not utilise artificial intelligence and cannot evaluate the appropriateness of treatment or care provided. However the PSR has also indicated that they have looked at the possibility of using artificial intelligence to assess medical records.

That does raise some interesting questions, both in the context of a push to expand Medicare to include more dental services and the use of artificial intelligence and big data more broadly to inform clinical practice guidelines.

Artificial intelligence is already being used to assist in systematic reviews and meta-analyses, which are crucial for developing evidence-based clinical guidelines. Machine learning algorithms can help screen large volumes of literature, extract relevant data, and identify patterns across studies. Many researchers are now moving to using tools access clinical records to ‘provide health services and researchers a valuable data source to monitoring health service utilisation, contribute to the evidence base through research and develop clinical decision support systems to improve quality of care.’ Given the lack of clear evidence-based guidelines in dentistry, there is an emerging opportunity to leverage the data that already exists in dental records to better inform clinical practice.

With ongoing calls to expand Medicare to include more dental services, some have expressed concern that this would inevitably lead to fraud, rorting or over-servicing. They particularly point to issues that occurred during the Chronic Disease Dental Scheme, which were a combination of administrative non-compliance with an unfamiliar scheme and allegations of over-servicing. Artificial intelligence tools may well be developed to assist in compliance or pre-approval for complex procedures to reduce the risk of non-compliance or over-servicing. The use of any such tools must take into account ethical concerns about data privacy and appropriate use of personal health data.

It is easy to foresee a future where algorithms are used to mine dental data to help inform best practice based on large datasets of clinical and radiographic findings, and this could then be used to inform utilisation and compliance with publicly funded programs. Whether this is sufficient to allay practitioner and public concerns about over-servicing remains to be seen.