In the 5 years, 2004-2008, there were 2575 admissions following acute HI in the PICANet dataset. This total represents 4.4% of all non-cardiac surgical intensive care admissions to PICUs in England and Wales. These episodes of intensive care were managed in 27 PICUs and the patients were aged 8.5 (3.1-12.8) years, median (IQR). Two-thirds were boys, 4.9% had complicating seizures or status epilepticus, and 2.1% suffered a cardiac arrest or event requiring cardiopulmonary resuscitation. Injuries involving other parts of the body occurred in 9.4% of episodes. These included: upper or lower limb, 6%; thorax or lung, 1.6%; abdomen or multiple injuries 1.8%. Overall 239 children died (mortality 9.3%).
The calibration of the PIM model in child HI is not perfect (PIM and PIM2 Hosmer-Lemeshow test χ2
= 0.01) and χ2
= 0.01), respectively) and the data are not spread evenly across the range of predicted risk (). However, the quantitative measures of performance suggest that the model can be considered ‘good’ (see Methods section).21
Cox's calibration regression showed an intercept −0.52 (95% CI, −0.85 to −0.19) and slope 1.19 (95% CI, 1.07 to 1.32) with χ2
< 0.001) for PIM and, intercept −0.54 (95% CI, −0.86 to −0.22) and slope 1.19 (95% CI, 1.07 to 1.32) with χ2
<0.001) for PIM2. The test of discrimination -area under the receiver operating characteristic curve - was 0.83 (95% CI, 0.80 to 0.86) for both models.
Calibration plots for PIM. The two plots show the observed risk by expected risk in deciles of expected risk using a log scale for the two versions of PIM.
The Brier's score of overall fit was 0.05 in both models. The PIM2 model appears better in .
Volume of PICU HI and neurosurgical practice
The system of PICU HI practice in England and Wales is illustrated in . The upper panel shows caseload for each PICU in ascending order of size (units 1 to 27). The lower panel shows cumulative practice in percentage, which forms a smooth curve with no obvious step-up or sigmoid shape. The borders between the four quartiles, from lower- to upper-quartile, coincide with units seeing more than 20, more than 25, and more than 40 HI admissions per year, respectively. summarises specialist paediatric neurosurgical PICU practice by unit and grouped according to the divisions presented in : upper first quartile, group I (three PICUs); second quartile, group II (four PICUs); third quartile, group III (six PICUs); and, lower fourth quartile, group IV (14 PICUs). The numbering of the units is the same as in . It is evident that the majority of group IV units do not have significant volume of specialist paediatric neurosurgical practice admitted to the PICU. Group I units have the largest specialist paediatric neurosurgical PICU activity. The hierarchy in HI and paediatric neurosurgical PICU practice shown in is also reflected in the volume of overall PICU practice per year in these groups of units (p<0.05). When all non-cardiac admissions are considered the median (IQR) admissions per year to PICUs in groups I to IV were 786 (618 to 1232), 399 (318 to 696), 493 (337 to 514) and 296 (198 to 377), respectively.
Figure 2 England and Wales PICU head injury practice by unit. Upper panel: Practice ordered according to size from smallest to largest. Lower panel: Cumulative practice by unit presented as percentage. The dotted lines in the lower panel show the borders of each (more ...)
Specialist paediatric neurosurgical practice by PICU. Grouping of PICUs into quartiles defined in from upperfirst to lowest-fourth, groups I to IV. PICU numbering is the same as in .
PICU practice and mortality
During the management of head-injured children in the PICU, the more severe cases undergo MV and ICP monitoring; less severe cases are observed without any use of these interventions. In one-third (33.5%) of the episodes MV and ICP monitoring was used, in another half of the episodes (51.5%) MV alone was used, and in the remaining 15% observation without use of these interventions was undertaken. MV with ICP monitoring was used in older patients: median 10.9 (IQR 5.1-13.6) years versus 6.5 (2.2-12.1) years in those undergoing MV alone and 7.8 (2.9-12.6) years in those being observed without these interventions, p< 0.001.
In the risk-adjusted regression model there was no apparent effect of increasing individual PICU HI caseload on reducing mortality. There were significant effects for both intensive care intervention category (i.e. MV and ICP monitoring, MV alone, and observation without these interventions; p<0.001) and HI admissions by PICU grouping (i.e. group I-IV; p< 0.005) in the model. The odds ratio (OR) for death in the ICP monitoring category, in comparison to MV alone, was 2.08 (95% CI, 1.43 to 3.00;p< 0.0001). summarises the grouped PICU data (i.e. groups I-IV) by category of interventions for HI practice (MV with ICP monitoring, MV alone, observation alone). There was some variation in case mix between the groups: group IV undertook fewer episodes of ICP monitoring and groups I and II had fewer episodes of observation without intervention. The MV with ICP monitoring category cases stayed longer in the PICU (median 6 days, χ2 734, p < 0.0001) and a higher proportion received vasoactive drugs (70.9%, χ2 1090, p < 0.0001). The SMR in the whole series of 2575 admissions was 1.06 (95% CI, 0.93 to 1.20), which indicates that the whole system performs as expected by the PIM model. On further inspection of , a higher percentage of ICP monitoring patients died (13.2%) when compared with the percentage of death in the other categories (p<0.001). The SMR was greater than 1.00 in the ICP monitoring category (1.31, 95% CI 1.09 to 1.67; p = 0.004). This finding was due to group II and group IV data - SMR 1.58 (95% CI, 1.04 to 2.29; p = 0.02) and 1.83 (95% CI, 1.18 to 2.73; p = 0.004) respectively - since in the other two groups the lower limit of the 95% confidence interval for SMR was below 1.00. These statistics equate with up to 11 extra deaths in the 175 group II cases undergoing MV with ICP monitoring, or 14 extra deaths in the 117 group IV cases undergoing MV with ICP monitoring.
Summary of PICU group data by category of intensive care intervention for HI care
The relationship between volume and outcome is examined in the funnel plots in . The PIM2 risk-adjusted mortality in each quartile is displayed as a scatter plot and compared with the funnel plot (upper panel). Both PIM models have been examined in the analysis, but since PIM2 gives the more conservative results these are presented. In addition we examined the data for evidence of inter-unit transfer during the acute ictus and found only 8 instances where this occurred and none of these episodes end in death of the child. The figure shows that no sector lies outside the 99.8% limits. However, groups II (PICUs 21–24 in ) and IV (PICUs 1–14 in ) lie outside the 95% upper warning limit. Further scrutiny of individual PICUs within each quartile is displayed in the lower funnel plot. Each of the 20 PICUs with more than 40 cases in 2004-2008 is shown. No centre lies outside the 99.8% limits. Six of the PICUs had risk-adjusted mortality that was beyond the upper 2σ warning limit: 4 of 7 group IV PICUs and 1 unit in each of groups I and II. Half of the units in group III appeared to demonstrate evidence of possible best practice since their risk-adjusted mortalities were placed between the lower 95% and lower 99.8% limits.
Figure 4 Funnel plots showing risk-adjusted mortality rate displayed as a scatter plot. The horizontal line shows the predicted mean 8.6%. Dotted lines show the 99.8% and 95% (2σ) limits. Points I–IV in the upper panel use the same notation as (more ...)
and together indicate there is excess mortality in groups II and IV, which may be a particular problem in those undergoing ICP monitoring. One difficulty with this analysis is that it considers experience during 2004-2008 as a whole. There is no insight into whether the finding represents earlier rather than more recent experience, which would be less relevant when considering contemporary performance in a system of care. The half-cumulative-sum-risk-adjusted charts shown in displays performance with reference to PIM2 in the case series for each of the four groups. is the control chart for group IV PICUs with plots for each episode in date and time sequence 2004-2008. Group IV PICUs have unexpectedly high number of deaths (upper red line in crossing upper threshold after ~550 cases). This finding is consistent with the observations in and and suggests that the finding is relevant to more contemporary practice. (group III PICUs, average annual HI caseload 20–25) signals an unexpectedly low number of deaths after every 200 patients (lower blue line crossing lower threshold after 200, 400, and 600 cases). The trend is also continued after case 600. Better than expected performance is also signalled in , but not with the same appearance in the plots. In (group II PICUs, average annual HI caseload 20-35) there is only one signal for unexpected low number of deaths (after ~500 cases) and the upper line crosses the 95% threshold between cases 300 and 400 in the sequence. In (group I PICUs, average annual HI caseload 4 40) there are three signals for unexpectedly low number of deaths, but the cycle length for the most recent signal is ~300 cases, which is greater than the cycle length in .
Figure 5 Half-cumulative-sum-risk-adjusted charts in the four quartiles of PICUs over their respective case series sequence, where: A, units 1–14, lower quartile and Group IV; B, units 15–20, third quartile and Group III; C, units 21–24, (more ...)