Of the 13,218 shifts in the dataset, 4,084 (31%) had missing data for at least one of the three variables of interest. The majority of these (23%) had a missing value for ‘care left undone’ or ‘neither agreed or disagreed/chose not to say. A further 226 shifts with understaffing less than zero, seven shifts with Agency ratio greater than one, and 60 with Agency ratio equal to one, were excluded from analysis due to concerns over data quality. Therefore, data from 8,841 shifts were included in the analysis.
Prevalence of care left undone by Setting
An important area of interest was for us to identify those clinical areas experiencing the highest prevalence of ‘care left undone’. The proportion of our responses by setting were; 15.3% from ED, 53.9% Adult Acute ward, 14% Critical Care, 11.2% were Older People’s ward and 5.6% Theatre. Within these responses, the highest proportion of “care left undone” was within the ED setting (48.4%) and lowest proportion within Theatre setting (21%, N = 122) (See Table 1). Adult acute ward reported, ‘care left undone’ on 45.3% of shifts (N = 2530), Older People’s Ward 46% (N = 535), Critical Care; 27.7% (N = 401).
Table 1
Staffing Levels and Self-reported Care Left Undone by Setting
| Care Left Undone |
Setting | Care Left Undone N (%) | No Care Left Undone N (%) | Neither Agree or Disagree | Total cases(N) |
Emergency Department/ Urgent and Emergency Care | 763 (48.4%) | 590 (37.3%) | 225 (14.3) | 1578 |
Adult Acute Ward | 2530 (45.3%) | 2244 (40.3%) | 805 (14.4%) | 5579 |
Critical Care/ high dependency | 401 (27.7%) | 854 (58.9%) | 194 (13.4%) | 1449 |
Older people’s ward | 535 (46%) | 438 (37.8%) | 188 (16.2%) | 1161 |
Theatre | 122 (21%) | 364 (63%) | 93 (16%) | 579 |
Having established that some clinical areas are vulnerable to ‘care left undone’, our next task was to focus on those resource related factors that may have contributed to this. Subsequently we ‘pooled’ data from all five clinical areas to ensure that the tests were adequately powered. Staffing ratio on shifts that reported ‘no care left undone’ was higher than on shifts that reported ‘care left undone’ Mdn = 0.94 (0.87–0.88) vs Mdn = 0.8 (0.79–0.8), p˂0.001 (Mann Whitney U test). There was a moderate, positive relationship between staffing ratio and care left undone rs (4942) = 0.25, p ˂0.05.
Further, we explored the relationship between the permanent staff ratio and care left undone on full complement shifts. The proportion of permanent staff was higher on shifts that reported ‘no care left undone’ (Mdn = 0.94 (IQR 0.8–0.82) than on shifts that reported ‘care left undone’ Mdn = 0.8 (IQR 0.72–0.74), p ˂ 0.05. Spearman’s rho correlation was carried out to assess the relationship between proportion of permanent staff and care left undone. There was significant evidence of a moderate correlation between proportion of permanent staff and care left undone on full complement shifts rs(8816) = 0.5, p < 0.01).
Furthermore, we were interested in the prevalence of staffing scenarios within our sample, particularly the occurrence of full complement shifts (shifts that fulfilled their planned quota of RNs), and shifts with at least one agency RN staff or more. Of the 8,841 shifts analysed, 3,449 (39%) had full complement (Understaffing = 0). 6,285 shifts (71%) had no agency staff (Agency ratio = 0), whilst 2,556 shifts (29%) had zero values for both predictors (see Table 2). Out of the full complement shifts that have agency RN’s, the most prevalent staffing scenarios were those with a proportion of agency (0.1–0.2, (N = 193), (0.3–0.4, N = 150) and 0.4–0.5 N = 196). In the low-understaffing category (25 per cent or less), the most prevalent staffing scenarios were those with a 0.1–0.2 and 0.2–0.3 proportion of agency staff. In the high-understaffing category (25% or more), the most prevalent staffing scenarios were those with a 0.3–0.4 (N = 306) and 0.4–0.5 (N = 202) proportion of agency staff (see Table 2).
Table 2
Number of Shifts by Category of Agency and Understaffing Ratio All Settings
| Understaffing Category |
Proportion of Agency Staff | 0 (Full complement) | 0.01–0.24 | ≥ 25 | Total |
0 | 2556 (28.9%) | 1519 (17.2) | 2210( 25) | 6285(71.1) |
0.001-0.1 | 56(0.6) | 143(1.6) | 10(0.1) | 209(2.36) |
(0.1–0.2) | 193(2.2) | 267 (3) | 107(1.2) | 567(6.4) |
(0.2,0.3) | 150(1.7) | 221(2.5) | 99(1.1) | 470(5.3) |
(0.3,0.4) | 220(2.5) | 87(1) | 306(3.5) | 613(7) |
(0.4,0.5) | 196(2.2) | 103(1.2) | 202(2.3) | 501(5.7) |
(0.5,0.6) | 12(0.1) | 18(0.2) | 9(0.1) | 39(0.4) |
(0,6, 0.7) | 35(0.4) | 9 (0.1) | 82(0.9) | 126(1.4) |
(0.7,1) | 9(0.1) | 12 (0.1) | 10(0.1) | 31(0.3) |
In addition, we were interested in the relationship between the occurrence of any ‘care left undone’ events during the shift (a binary outcome) and two predictor variables, namely “Understaffing” (1 minus the ratio of actual to planned number of registered nurses) and “Agency ratio” (ratio of agency to total actual registered nurses working the shift). Table 3 presents the estimates and 95% confidence intervals for the probability of a ‘care left undone’ event at a range of values for Agency ratio in three grouped categories of understaffing ratio: 0, 0.01–0.24, and > = 0.25. The data was categorised based on our interest in shifts that had no understaffing (understaffing = 0), low level of understaffing that happens due to unforeseen circumstances like for instance sick leave. Understaffing of 25% or more was considered moderate to high level of understaffing that might have significant impact on patient care. These categories each account for roughly a third of the shifts (N = 3427, 2379 and 3035 respectively). The odds calculations are based on the smoothed regression shown in Fig. 1, which presents this relationship graphically.
Table 3
Estimated odds of any missed care event, by agency ratio for three Understaffing categories
Odds of Care Left Undone, (95% CI) |
| Understaffing category |
Agency Ratio | 0 | 0.01–0.24 | ≥ 0.25 |
0 | 0.50 (0.46, 0.54) | 0.91 (0.82, 1.00) | 1.65 (1.52, 1.80) |
0.1 | 0.53 (0.48, 0.59) | 1.06 (0.96, 1.17) | 1.91 (1.73, 2.10) |
0.2 | 0.57 (0.48, 0.67) | 1.24 (1.05, 1.45)* | 2.19 (1.90, 2.53)* |
0.3 | 0.63 (0.52, 0.75) | 1.42 (1.15, 1.76)* | 2.52 (2.08, 3.06)* |
0.4 | 0.74 (0.62, 0.90)* | 1.52 (1.19, 1.94)* | 2.72 (2.24, 3.30)* |
0.5 | 1.00 (0.79, 1.25)* | 1.66 (1.21, 2.27)* | 2.83 (2.23, 3.59) |
0.6 | 1.49 (1.08, 2.06)* | 1.97 (1.27, 3.04)* | 2.85 (2.12, 3.83) |
0.7 | 2.23 (1.37, 3.64)* | 2.34 (1.22, 4.48)* | 2.81 (1.82, 4.32) |
*Statistically significant odds ratio difference when comparing to the full complement shifts with no agency staff (p < 0.05). |
The trend of odds of ‘care left undone’ in Fig. 1 shows that the odds of care left undone increases, with varying amounts, on shifts where there is increasing reliance on temporary staff, as indicated by increasing proportions of agency staff.
[Insert Figure 1 here.]
Our findings demonstrate that on full complement shifts with greater proportions of agency staff, the odds of care left undone increase. For instance, when comparing shifts that have no agency staff (Odds = 0.5, 95% CI, 0.46–0.54) to when the proportion of agency staff is 20 per cent of the staffing (Odds = 0.57, 95% CI, 0.48–0.67), we see that the odds of missed care increase by 14% (OR = 1.14, 95% CI, 1.04–1.24). This difference becomes statistically significant on full complement shifts with 40 per cent or more agency staff OR = 1.48, 95% CI, 1.4–1.7), (p < 0.05).
Similarly, where there is a 20 per cent reliance on agency staff, and RN understaffing is 25 per cent or less, the odds of ‘care left undone’ is 1.24 (1.05–1.45). The odds of ‘care left undone’ when agency usage is 20 per cent rises where the RN staff ratio is less than 75 per cent of planned full complement to 2.19 (1.9–2.53), p < 0.05.
Furthermore, considering a level of 25 per cent or more understaffing, the odds of missed care events increase further still but only slightly with an increase in proportion of agency staff. Relating these results back to Fig. 1, whilst it can be observed that, within the range of 0–24 per cent of understaffing, there is a steady increase in the occurrence of ‘care left undone’, relatively little further increase is observed in the 25 per cent or more range. A simple linear logistic regression model confirms that the increase in missed care probability is statistically significant (p < 0.001) for both predictors separately, but the addition of significant quadratic and interaction terms confirms that the simple model does not adequately describe the patterns in the data.