We have recently demonstrated that the distribution of total time spent by patients in emergency departments (EDs) in England shows a peak immediately prior to the current Department of Health target of 4 hours.¹ We aimed to investigate whether this phenomenon suggested that performance data were being manipulated. We hypothesised that if data regarding the total length of time patients spend in emergency department were being manipulated, then such data would show evidence of digit preference bias (DPB). DPB commonly occurs in the manual recording of numerical data.^2–4 While it may be expected to occur to some extent in the recording of total time in the ED, the degree of DPB would not be expected to change over different time points. Such methods of detecting data manipulation have been recently highlighted by Al-Marzouki et al.⁵

METHODS, RESULTS, AND DISCUSSION

The methods of data collection have been described previously.¹ We have now collected data from 117 EDs in England representing 621 674 patient episodes occurring during April 2004. Of these episodes, 616 067 (99.1%) had valid times of arrival and departure and were included in the analysis. The total time in the ED, defined as time from arrival until departure, was calculated for each patient episode, and the terminal digit of the total time in minutes was recorded. Episodes were grouped according to total time in the ED into 10 minute blocks (0–9, 10–19, etc). The percentage of episodes in each block with a particular terminal digit was determined.

As can be seen from fig 1, the percentage of episodes with a terminal digit of 0 or 9 changed little for the periods preceding and following the target time of 240 minutes. The mean proportion of episodes with a terminal digit of 0 or 9 was 10.7% and 10.0% respectively, for the groups with a duration of between 0 and 229 minutes. Of the episodes with a duration of 230–239 minutes, 8.7% (n = 1915) had a terminal digit of 0, and 14.7% (n = 3215) a terminal digit of 9. For episodes with a duration of 240–249 minutes, these figures are 31.8% (n = 1434) and 7.3% (n = 329), respectively. The proportion of episodes with durations between 230 and 249 minutes that have other terminal digits is shown in table 1.

There were 21 927 episodes with a duration of 230–239 minutes. Assuming that the distribution of terminal digits should be even from 0 to 9, it would be expected that 10% (n = 2193) of episodes would have a terminal digit of 9, therefore 1024 more episodes than expected had a total time in the ED of 239 minutes. There were 4505 episodes with a duration of 240–249 minutes, therefore 987 more episodes than expected had a duration of 240 minutes. In total, evidence of manipulation of performance data appeared to be present in up to 2011 (0.3%) episodes.

To determine the extent to which DPB was present in individual EDs, the percentage of episodes with a duration of 230–239 minutes and a terminal digit of 9 was determined for each ED. Only those EDs with a least 20 episodes occurring during this period were included (n = 114). The median percentage was 11.1% (interquartile range (IQR) 4.9%) with a maximum of 57.3%. The same analysis was applied to episodes with a duration of 240–249 minutes and a terminal digit of 0. Again, only EDs with at least 20 episodes occurring during this period were included (n = 85). The median percentage with a terminal digit of 0 was 22.9% (IQR 29.7%, maximum 97.6%), with five EDs having in excess of 80% of episodes with a terminal digit of 0 during this period.

CONCLUSION

The data presented here show that considerable DPB exists in the total time in ED for episodes with a duration between 230 and 249 minutes. It is likely that the true performance of some EDs in relation to the 4 hour target is considerably lower than reported.

DPB appears to affect only a small proportion of episodes. However, during the period April 2004 to March 2005 there were 16.71 million attendances at EDs and minor injury units in England.⁶ If DPB affects the recording of total time in the ED in all of these units it would affect over 50 000 episodes per year, equivalent to the total annual attendance of a medium sized ED.