According to findings published in the American Journal of Roentgenology, artificial intelligence predictive analytics performs reasonably well in solving complex operational problems – outpatient MRI appointment no-shows, in particular – using a modest amount of data and simple function engineering, and may help minimise such no-shows.
What is easy and advantageous about the data is that it is easily retrievable from frontline IT systems which are widely used in hospital radiology departments in many cases. It can also be easily incorporated into regular workflows, which the authors said may enhance healthcare delivery quality and effectiveness.
WHAT’S THE IMPACT?
In order to train and validate this model, the researchers team collected records of 32,957 outpatient MRI appointments scheduled from their institution’s radiology information system between January 2016 and December 2018, while obtaining a further 1,080 record holdout test set from January 2019. The no-show rate was 17.4 per cent overall.
After testing various predictive models of machine learning built with commonly used, open-source software tools, the team implemented a decision-based ensemble algorithm using a gradient boosting framework: XGBoost, version 0.80.
Approximately converted, this culminated in an intervention measure of phone alerts for patients with the top 25 percent highest probability of a no-show appointment, which was carried out over six months.
Six months after implementation, the predictive model’s no-show rate was 15.9 percent , compared to 19.3 percent in the corresponding 12-month pre-intervention period – a 17.2 percent increase from the no-show baseline rate. The no-show rates of touch and non-contact patients in the group with high risk of appointment no-shows as expected by the model were 17.5% and 40.3% respectively.
The aim was not to produce a complex model, but rather a simple one that could be developed quickly, with minimal data processing.
THE LARGER TREND
Data released in 2016 found that missed and open appointments cost $150 billion a year for the healthcare industry. Providers have national no-show rates ranging from 5 per cent to 30%. Currently any 60-minute open or no-show slot costs doctors $200.
A 2018 MGMA Stat poll found that texting is the most preferred and efficient tool when it comes to cutting down on patient no-shows and staff time spent confirming appointments. Most medical groups said they were already engaging with their patients by text, and many said they were using an alternative such as notifications for email and phone calls. Source