Introduction As physiologists we spend a vast amount of time manually analysing sleep studies. As technology advances, we are often told that the auto analysis programmes are improving with every update. We decided to find out how accurate it has become and whether it can be relied on for an accurate clinical outcome.

Methods We utilised clinical downtime during the COVID-19 pandemic to carry out a service audit between September 2019 and March 2020. 160 studies were reviewed with a comparison between automatic and manually scored AHI. Pearsons co-efficient of variation and Bland-Altman analysis was carried out both to investigate the correlation of the assays as well as allowing clinical judgement into the significance.

Results We found that 90 results changed scoring classification and 70 stayed the same. A Pearsons co-efficient of variation was calculated at 0.86 indicating a very high correlation (Schober, Boer and Schwarte, 2018), however overall, we found an average AHI difference of 8.4 in the scores between auto and manual. Looking closely at the distribution on the plot it suggests that the auto analysis programme correlates highly when the AHI is < 7 with a tightly packed area of datum however as the AHI increases so does the scatter.

Discussion We found although there is a positive correlation auto analysis is unable to operate in a small enough window to have no impact on the treatment pathway. Relying on an automatic programme would indeed ‘speed up’ the service but at the cost of accurate clinical science as well as physical cost on misdiagnosis.

We are aware this study is limited to the NOX T3 device and a relatively small data set so may not extend beyond these limitations.

Reference

1. Schober P, Boer C, and Schwarte L. Correlation Coefficients. Anesthesia & Analgesia 2018;126(5):1763-1768.