When thinking about data-quality we often take this to simply mean the completeness of that data. This is usually because reporting is only concerned with the amount of data, rather than whether it accurately reflects events.

The completeness of a data item is important and this is why you will often find systems force you to state ‘not recorded’ rather than leaving an item blank. However, just because something is recorded it does not necessarily mean it was accurate.

How can this be assessed without going back in time to check what was measured reflects what was recorded? Whilst this is not possible for individual data points, it can be assessed when looking at the data as a whole, specifically the distribution of values within that data.

One example of this is digit preference bias. This term describes the tendency for people to record numbers ending in particular digits, often 0 or 5. For example, in healthcare data this has been observed in relation to recording blood pressures, where historically it was common to record values such as ‘120/70’, especially using manual sphygmomanometers.

The phenomenon has also been demonstrated in operational data, such as when time-points are recorded.

The graph above shows the recorded minute of arrival and departure for 72,000 emergency department attendances. Time of arrival was recorded automatically at registration and shows an even distribution across the values from 0 to 59.

The time of departure was recorded on paper by clinical staff. It is clear that there is a preference of values ending in 0 or 5, this being particularly prominent at 0, and 30 minutes. This cannot be an accurate representation of when patients left a department.

Techniques such as this can be used to determine if data in your systems is being recorded accurately.

Digit preference bias in the recording of emergency department times. TE Locker, SM Mason. European Journal of Emergency Medicine 13 (2), 99-101