- Email: [email protected]
Comparison of UK paediatric cardiac surgical performance by analysis of routinely collected data 1984–96: was Bristol an outlier?
ARTICLES
Comparison of UK paediatric cardiac surgical performance by analysis of routinely collected data 1984–96: was Bristol an outlier? Paul Aylin, Bernadette Alves, Nicky Best, Adrian Cook, Paul Elliott, Stephen J W Evans, Audrey E Lawrence, Gordon D Murray, John Pollock, David Spiegelhalter
Summary Background Reports of high mortality after paediatric cardiac surgery at the Bristol Royal Infirmary, UK, led to the establishment of an independent public inquiry. A key question was whether or not the mortality statistics in Bristol were unusual compared with other specialist centres. To answer this question, we did a retrospective analysis of mortality in the UK using two datasets. Methods Data from the UK Cardiac Surgical Register (CSR; January, 1984, to March, 1996) and Hospital Episode Statistics (HES; April, 1991, to December, 1995) were obtained for all 12 major centres in which paediatric cardiac surgery is done in the UK. The main outcome measure was mortality within 30 days of a cardiac surgical procedure. We estimated excess deaths in Bristol using a random-effects model derived from the remaining 11 centres. Additionally, a sensitivity analysis was done and case-mix examined. Findings For children younger than 1 year, in open operations, the mortality rate in Bristol was around double that of the other centres during 1991–95: within the CSR, there were 19·0 excess deaths (95% interval 2–32) among 43 deaths; and in HES, there were 24·1 excess deaths (12–34) among 41 deaths recorded. There was no strong evidence for excess mortality in Bristol for closed operations or for open operations in children older than 1 year. Interpretation Our results suggest that Bristol was an outlier, and we do not believe that statistical variation, systematic bias in data collection, case-mix, or data quality can explain a divergence in performance of this size. Lancet 2001; 358: 181–87
Division of Primary Care and Population Health Sciences, Imperial College School of Medicine, London, UK (P Aylin FFPHM, B Alves BA, N Best PhD, A Cook MSc, Prof P Elliott FRCP); Medicines Control Agency, London (Prof S J W Evans MSc); Lawrence Research, Bishopton (A E Lawrence MSc); Department of Community Health Sciences, University of Edinburgh Medical School, Edinburgh (Prof G D Murray PhD, J Pollock BSc); and MRC Biostatistics Unit, Institute of Public Health, Cambridge (D Spiegelhalter PhD) Correspondence to: Dr Paul Aylin, Department of Epidemiology and Public Health, Imperial College School of Medicine, St Mary’s Campus, Norfolk Place, London W2 1PG, UK (e-mail: [email protected])
THE LANCET • Vol 358 • July 21, 2001
Introduction Reports of high mortality within paediatric cardiac surgery units at the Bristol Royal Infirmary, UK, led to the establishment of an independent public inquiry.1 One of the key questions was whether or not the mortality statistics in Bristol were unusual compared with other specialist centres. This paper presents an analysis of data on paediatric cardiac surgery carried out at Bristol and 11 other major centres in England. We have previously shown that, within Bristol, levels of activity (numbers of cases and deaths) recorded in seven available datasets held for purposes of clinical audit or management (and, in one case, collated specifically for the Inquiry) were broadly similar, with no large discrepancies that could not be accounted for.2–6 Of these datasets, two were national: Hospital Episode Statistics (HES) and the UK Cardiac Surgical Register (CSR). Despite the limitations of routine data, national comparisons with these two sets of data could be undertaken. Although our study concerns outcome of paediatric cardiac surgery, its approach and methodology is applicable more generally to comparison of performance across a wide range of specialties.
Methods Data from the CSR (January, 1984, to March, 1996) and HES (April, 1991, to December, 1995) were obtained for all 12 major centres in England in which paediatric cardiac surgery is done. Two age-groups were defined: children younger than 1 year, and children aged between 1 and 15 years. Two different classifications of operative procedures were used: the first comprised broad classes of either open (requiring cardiopulmonary bypass) or closed operations, and the second consisted of 13 procedure groupings (11 open, two closed).7 Because the two data sources were available for different times within the study period, four epochs were defined:2 (1) January, 1984, to December, 1987; (2) January, 1988, to December, 1990; (3) January, 1991, to March, 1995; and (4) April, 1995, to December, 1995. For the HES data, mortality rate was based on admissions for whom discharge status was known (discharge home, transfer to another hospital, or death in hospital). Admissions with unknown discharge status were excluded. Our main outcome measure was mortality within 30 days of the procedure. Three stages of analysis were undertaken. First, overall mortality within each epoch was estimated with 95% CIs, and centres were ranked with 95% intervals for each rank.8 Second, expected mortality in Bristol was estimated, allowing for betweencentre variability. Specifically, within each procedure group we modelled between-centre variability using a variance-components (random-effect) model9 for the 11 centres (excluding Bristol), in which mortality rates (on a logit scale) were assumed drawn from a normal population distribution:10 the precisions of these normal
181
For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
distributions were further allowed to vary between procedure groups. The expected number of deaths in a hypothetical centre with Bristol’s level of activity was then predicted,6 and excess deaths in Bristol were estimated by subtracting these expected numbers from the observed numbers of deaths. The entire analysis was then repeated in turn for each of the remaining centres. The third stage comprised a traditional analysis to assess statistical significance and its robustness to alternative assumptions. A sensitivity analysis used four scenarios that contrasted mortality at Bristol with the average of the mortality rates in the other centres, by use of a logistic regression model to adjust for differences between centre and the mix of open surgical procedures. In the first scenario, we excluded from the analysis three centres for which comparison of the two data sources gave more than 20% difference in death rates; in the second, we removed procedure groups 2, 3, and 5 (interatrial and other repairs to correct transposition of
great arteries, and repair of complete atrial ventricular septal defects), which were prominent contributors to the observed overall excess mortality; in the third, we corrected undercount in the HES for all deaths within 30 days, estimated from a mortality linkage study,11 and recalculated the mortality rates for each centre; and in the fourth scenario, for each procedure group, we selected either HES or CSR data depending on which had the lower mortality for Bristol, and then, for the remaining centres, we chose the data source which had the higher mortality. Two-sided p values of less than 0·004 were sought as a Bonferroni adjustment for multiple comparisons. To examine possible confounding by case-mix, we obtained from HES: information on age at operation, proportions of children with a mention of Downs’ syndrome in any diagnosis field, the proportion of transfers from other units, and the proportions of cases admitted as an emergency, and compared mortality by
CSR data 1984–87, younger than 1 year
CSR data 1988–90, younger than 1 year
CSR data 1991–Mar 95, younger than 1 year
HES data 1991–Mar 95, younger than 1 year
Centre Mortality
Centre Mortality Approximate 95% interval
Centre Mortality Approximate 95% interval
Centre Mortality Approximate 95% interval
1 31/108 (29%) 2 22/107 (21%) 3 35/135 (26%) 4 14/45 (31%) 5 26/104 (25%) 6 34/198 (17%) 7 25/184 (14%) 8 57/362 (16%) 9 11/79 (14%) 10 34/90 (38%) 11 57/438 (13%) 12 21/121 (17%)
1 43/181 (24%) 2 27/200 (13%) 3 26/157 (17%) 4 15/142 (11%) 5 36/217 (17%) 6 49/417 (12%) 7 27/253 (11%) 8 57/369 (15%) 9 28/214 (13%) 10 31/184 (17%) 11 67/740 (9%) 12 32/268 (12%)
1 41/143 (29%) 2 25/187 (13%) 3 24/323 (7%) 4 23/122 (19%) 5 25/164 (15%) 6 42/405 (10%) 7 24/239 (10%) 8 53/482 (11%) 9 26/195 (13%) 10 25/177 (14%) 11 58/581 (10%) 12 31/301 (10%)
Total 367/1971 (19%)
Total 438/3342 (13%)
Total 397/3319 (12%)
CSR data 1988–90, 1–15 years
CSR data 1991–Mar 95, 1–15 years
HES data 1991–Mar 95, 1–15 years
Centre Mortality Approximate 95% interval
Centre Mortality Approximate 95% interval
Centre Mortality Approximate 95% interval
Centre Mortality Approximate 95% interval
1 24/284 (8%) 2 8/203 3 17/219 (8%) 4 0/15 (0%) 5 21/233 (9%) 6 49/567 (9%) 7 11/258 (4%) 8 52/459 (11%) 9 17/155 (11%) 10 32/213 (15%) 11 16/202 (8%) 12 14/380 (4%)
1 37/304 (12%) 2 10/132 (8%) 3 27/288 (9%) 4 3/107 (3%) 5 16/186 (9%) 6 44/397 (11%) 7 8/238 (3%) 8 37/776 (5%) 9 7/164 (4%) 10 28/161 (17%) 11 25/570 (4%) 12 19/300 (6%)
1 28/382 (7%) 2 6/343 (2%) 3 17/199 (9%) 4 9/67 (13%) 5 20/330 (6%) 6 33/623 (5%) 7 9/359 (3%) 8 18/525 (3%) 9 12/345 (3%) 10 48/373 (13%) 11 25/723 (3%) 12 35/616 (6%)
1 21/314 (7%) 2 5/216 (2%) 3 20/406 (5%) 4 6/168 (4%) 5 10/270 (4%) 6 27/738 (4%) 7 6/249 (2%) 8 32/642 (5%) 9 14/232 (6%) 10 35/292 (12%) 11 14/527 (3%) 12 25/471 (5%)
Total 261/3188 (8%)
Total 261/3623 (7%)
Total 260/4885 (5%)
Total 215/4525 (5%)
Approximate 95% interval
1 16/63 (25%) 2 11/66 (17%) 3 10/36 (28%) 4 0/0 (0%) 5 23/83 (28%) 6 48/242 (20%) 7 19/186 (10%) 8 55/236 (23%) 9 15/68 (22%) 10 28/109 (26%) 11 30/77 (39%) 12 28/187 (15%)
Total 283/1353 (21%)
CSR data 1984–87, 1–15 years
0 10 20 30 40
0 10 20 30 40
0 10 20 30 40
0 10 20 30 40
Surgical mortality rate (%)
Figure 1: Mortality from open procedures by centre derived from Hospital Episode Statistics (HES) and Cardiac Surgical Register (CSR) Centre 1=Bristol Royal Infirmary. HES mortality rates based on admissions with known outcome.
182
THE LANCET • Vol 358 • July 21, 2001
For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
Type of operation and data source
Epoch
All open operations CSR 1 (1984–87) 2 (1988–90) 3 (1991–95) 4 (1995–96)* Total (1984–96) HES 3 (1991–Mar 1995) 4 (1995)* Total (1991–95)
Children younger than 1 year
Children aged 1–15 years
Mortality in Bristol (deaths/cases)
Mortality elsewhere (deaths/cases)
Expected Excess deaths
Mortality in Bristol (deaths/cases)
Mortality elsewhere (deaths/cases)
Expected deaths
Excess
16/63 (25%) 31/108 (29%) 43/181 (24%) 3/50 (6%) 93/402 (23%) 41/143 (29%) 2/24 (8%) 43/167 (26%)
275/1308 (21%) 336/1863 (18%) 395/3161 (12%) 126/1049 (12%) 1132/7381 (15%) 356/3176 (11%) 68/563 (12%) 424/3739 (11%)
14·0 22·3 24·0 6·0 66·3 16·9 2·8 19·7
2·0 8·7 19·0 3·0 26·7 24·1 1·8 22·3
24/284 (8%) 37/304 (12%) 28/382 (7%) 2/136 (1%) 91/1106 (8%) 21/314 (7%) 0/87 21/401 (5%)
242/2989 (8%) 225/3333 (7%) 232/4508 (5%) 42/1305 (3%) 741/12 135 (6%) 194/4211 (5%) 31/695 (4%) 225/4906 (5%)
23·3 22·4 22·8 4·4 72·9 15·0 3·7 18·7
0·7 14·6 5·2 2·4 18·1 6·0 3·7 2·3
112/1851 (6%) 96/1750 (5%) 57/1839 (3%) 18/658 (3%) 283/6098 (5%) 78/1784 (4%) 25/357 (9%) 103/2141 (5%)
9·4 7·9 6·2 1·5 25·0 6·9 2·8 9·7
8·6 4·1 1·2 1·5 10·0 0·1 2·8 2·7
3/120 (2%) 4/127 (3%) 3/88 (3%) 1/24 (4%) 11/359 (3%) 0/89 1/28 (4%) 1/117 (1%)
21/1293 (2%) 21/1002 (2%) 21/792 (3%) 3/233 (1%) 66/3320 (2%) 15/893 (2%) 0/111 15/1004 (1%)
2·0 2·6 2·5 0·3 7·4 1·7 0 1·7
1·0 1·4 0·5 0·7 3·6 1·7 1·0 0·7
13·0 19·0 17·1 2·4 51·5 14·8 2·7 17·5
3·0 7·0 12·9 0·4 22·5 27·2 1·7 25·5
13·7 12·7 12·2 1·7 40·3 12·0 1·7 13·7
2·3 11·3 2·8 1·7 14·7 4·0 1·7 2·3
All closed operations CSR 1 (1984–87) 18/154 (12%) 2 (1988–90) 12/152 (8%) 3 (1991–95) 5/179 (3%) 4 (1995–96)* 0/54 Total (1984–96) 35/539 (6%) HES 3 (1991–Mar 1995) 7/153 (5%) 4 (1995)* 0/31 Total (1991–95) 7/184 (4%) 11 open procedure groups (adjusted for groupings) CSR 1 (1984–87) .. 2 (1988–90) .. 3 (1991–95) .. 4 (1995–96)* .. Total (1984–96) .. HES 3 (1991–Mar 1995) .. 4 (1995)* .. Total (1991–95) ..
.. .. .. .. .. .. .. ..
.. .. .. .. .. .. .. ..
.. .. .. .. .. .. .. ..
CSR=Cardiac Surgical Register. HES=Hospital Episode Statistics. HES mortality rates based on admissions with known outcomes. *Simplified analysis, excluding random effects.
Table 1: Observed and expected number of deaths and excess mortality in Bristol relative to other 11 centres in England
levels of socioeconomic deprivation (obtained from the Carstairs deprivation score12 for census enumeration districts linked to postcodes of residence). For open operations, the number and mortality rate at each month of age was calculated for each centre.
the third highest mortality rate (95% interval 1–6) reported to the CSR during 1988–90, and the highest rate in both the CSR (1–2) and HES (1–1) during 1991–95. Only one other centre, centre 10 (Harefield hospital), showed consistent evidence across the two sources of data of divergent performance, and this was for open operations in children older than 1 year. For closed procedures (not shown), no centres showed consistent evidence of divergent performance. Table 1 summarises mortality results for open and closed classes of procedures and the 11 open procedure
Results Figure 1 shows mortality by centre for open operations. Between-centre variability in outcomes reported to the CSR declined over the three epochs. In children younger than 1 year, out of the 12 centres, Bristol (centre 1) had Procedure
Mortality according to CSR Bristol
Open procedure groups G1 Tetralogy of Fallot 0/2 (0%) G2 Interatrial TGA 0/3 (0%)* G3 Other TGAs 10/36 (28%)* (switch) G4 Repair of TAPVD 6/18 (33%) G5 Repair of AVSD 8/32 (25%) (complete not partial) G6 Closure of secundum 2/5 (40%) and sinus venosus ASD G7 Closure of VSD 0/51 (0%) G8 Truncus arteriosus 2/7 (29%) G9 Fontan type 1/1 (100%) operations G10 Aortic, pulmonary valve, 1/1 (100%) and paravalve procedures G11 Mitral valve procedures 0/1 (0%) Closed procedure groups G12 Closed shunts G13 Coarctation procedures
0/30 (0%)
Elsewhere
Mortality according to HES ObExExcess served pected
Bristol
Elsewhere
Observed
Expected
Excess
22/270 (8%) 0 14/50 (28%)* 0 72/573 (13%)* 10
0·2 0·8 5
0·2 0·8 5
0/3 (0%) 2/15 (13%) 10/13 (77%)
18/281 (6%) 11/98 (11%) 60/573 (10%)
0 2 10
0·2 1·6 1·5
0·2 0·4 8·5
21/153 (14%) 44/331 (13%)
6 8
2·6 4·5
3·4 3·5
5/14 (36%) 11/23 (48%)
23/166 (14%) 48/402 (12%)
5 11
2 3
3 8·0
2/129 (2%)
2
0·1
1·9
5/10 (50%)
11/149 (7%)
5
0·7
4·3
18/689 (3%) 26/102 (25%) 7/21 (33%)
0 2 1
1·4 1·9 0·4
1·4 0·1 0·6
0/47 (0%) 3/4 (75%) 2/4 (50%)
45/811 (6%) 31/97 (32%) 20/118 (17%)
0 3 2
2·7 1·3 0·7
2·7 1·7 1·3
24/179 (13%)
1
0·2
0·8
2/4 (50%)
25/250 (10%)
2
0·4
1·6
5/36 (14%)
0
0·2
0·2
2/3 (67%)
12/51 (24%)
2
0·7
1·3
9/464 (2%)
0
0·6
0·6
3/38 (8%) 2/64 (3%)
58/598 (10%) 17/426 (4%)
3 2
3·8 2·6
0·8 0·6
CSR=Cardiac Surgical Register. HES=Hospital Episode Statistics. HES mortality rates based on admissions with known outcomes. ASD=atrial septal defect. AVSD=atrioventricular septal defect. TAPVD=total anomalous pulmonary-venous drainage. TGA=transposition of great arteries. VSD=ventricular septal defect. *Data are unreliable since seems to have been misuse of these groupings for operations such as “Mustard” and “Senning”.
Table 2: Observed and expected mortality relative to other 11 centres in England: 11 open and two closed procedure groups on children younger than 1 year during epoch 3 (April, 1991, to March, 1995)
THE LANCET • Vol 358 • July 21, 2001
183
For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
Data source and epoch
CSR 2 (1988–90)
3 (1991–95)
HES 3 (1991–95)
HES+CSR 3 (1991–95)
Analysis
Unadjusted for procedure grouping
Adjusted for procedure grouping
Odds ratio* (95% interval)
p
Odds ratio* (95% interval)
p
Baseline 1 2 Baseline 1 2
1·54 (1·03–2·30) 1·73 (1·01–2·98) 2·38 (1·44–3·94) 1·95 (1·40–2·72) 1.98 (1·27–3·10) 1·96 (1·28–3·01)
0·04 0·04 0·001 0·0001 0·002 0·002
1·62 (0·96–2·72) 1·86 (0·93–3·73) 2·88 (1·50–5·53) 1·90 (1·22–2·95) 1·97 (1·09–3·56) 1·90 (0·95–3·82)
0·07 0·08 0·001 0·004 0·03 0·07
Baseline 1 2 3
2·72 (1·92–3·86) 2·74 (1·71–4·38) 1·80 (1·10–2·95) 2·67 (1·71–4·17)
0·0001 0·0001 0·02 0·0001
3·43 (2·21–5·32) 3·46 (1·85–6·48) 2·58 (1·46–4·58) 3·20 (1·47–6·95)
0·0001 0·0001 0·001 0·003
4
1·83 (1·31–2·54)
0·0003
1·46 (0·97–2·21)
0·07
Analysis 1 removes three centres with highest discrepancies between Hospital Episode Statistics (HES) and Cardiac Surgical Register (CSR) data. Analysis 2 removes procedures for transposition of great arteries and atrioventricular septal defect. Analysis 3 corrects for undercount in HES statistics for all deaths within 30 days. Analysis 4 selects HES or CSR data for each procedure group dependent on which has the lower mortality (if Bristol), or the higher mortality (if elsewhere). HES mortality rates based on admissions with known outcomes. *Estimated ratio of odds of dying in Bristol compared with average odds of dying in other centres.
Table 3: Odds ratios and sensitivity analysis for open operations on children younger than 1 year
groupings, with data from the CSR and HES, for children younger or older than 1 year in Bristol and the 11 other centres. At ages younger than 1 year, in open operations within the CSR, there were 19·0 excess deaths (95% interval 2–32) among 43 deaths between January, 1991, and March, 1995—a time when the mortality rate in Bristol was around double that in other centres (24 vs 12%). There was no strong evidence for excess mortality for open operations in children older than 1 year during the same period. Reported mortality for open operations in children younger than 1 year fell in other centres from 21% in epoch 1 to 12% in epoch 3, whereas it did not improve at Bristol. There was no evidence of excess mortality in Bristol during epoch 4, although activity in Bristol was too small to draw any firm conclusion. Within the closed class of procedures, there was no strong evidence for excess mortality. HES data confirmed the pattern of deaths between April, 1991, and March, 1995: there were 24·1 excess deaths (12–34) among 41 deaths in Bristol and a mortality of 29% versus 11% elsewhere. There was no
strong evidence of excess mortality in closed procedures or in open operations in children older than 1 year. Within individual procedure groups for epoch 3 (1991–95) for children younger than 1 year, excess mortality with 95% confidence was identified within HES data for repairs to correct transposition of great arteries (other than interatrial), repair of complete atrial ventricular septal defects, and closure of secundum and sinus venosus atrial septal defects (table 2). After adjustment for procedure group, excess mortality at Bristol was retained, with 12·9 excess deaths in the CSR during 1991–95, and 27·2 in HES for the same epoch (table 1). Over all other centres, the corresponding maximum figure was 5·3 for CSR (centre 10) and 9·0 for HES (centre 4). Table 3 shows odds ratios and their 95% CI for the odds of dying in Bristol compared with the average odds of dying in the other centres for the 11 open procedure groupings (unadjusted and adjusted for procedure groupings) for children younger than 1 year. For the CSR, odds ratios increased from 1·54 (95% CI
Bristol
Elsewhere
Number of admissions
Mortality* (95%CI)
Number of admissions Mortality* (95%CI)
Age <90 days 90 days–1 year 1–15 years
37 (7%) 135 (27%) 333 (66%)
19/30 (63% [44–80]) 22/113 (19% [13–28]) 21/314 (7% [4–10])
1696 (22%) 1641 (21%) 4408 (57%)
254/1623 (16% [14–18]) 102/1562 (7% [5–8]) 195/4293 (5% [4–5])
Down’s syndrome Mentioned in any diagnosis field Not mentioned
52 (10%) 453 (90%)
7/51 (14% [6–26]) 55/406 (14% [10–17])
539 (7%) 7206 (93%)
43/537 (8% [6–10]) 508/6941 (7% [7–8])
Transfers From other units Non-transfers
32 (6%) 473 (94%)
15/26 (58% [37–77]) 47/431 (11% [8–14])
1707 (22%) 6038 (78%)
236/1656 (14% [12–16]) 315/5822 (5% [5–6])
Emergencies Emergency admissions Non-emergency admissions
36 (7%) 469 (93%)
13/28 (46% [28–66]) 49/429 (11% [9–15])
783 (10%) 6962 (90%)
88/751 (12% [10–14]) 463/6727 (7% [6–7])
Carstairs quintile* 1 (least deprived) 2 3 4 5 (most deprived) Unknown
95 (19%) 105 (21%) 134 (27%) 102 (20%) 67 (13%) 2 (0%)
9/82 (11% [5–20]) 10/88 (11% [6–20]) 11/122 (9% [5–16]) 11/88 (13% [6–21]) 1/55 (2% [0–10]) 1/1 (100% [3–100])
1161 (15%) 1274 (16%) 1281 (17%) 1327 (17%) 1634 (21%) 1068 (14%)
69/1116 (6% [5–8]) 86/1241 (7% [6–8]) 78/1253 (6% [5–8]) 91/1290 (7% [6–9]) 112/1592 (7% [6–8]) 97/977 (10% [8–12])
HES mortality rates based on admissions with known outcome. *Based on postcode of residence, where recorded.
Table 4: Status at admission and comorbidity for open class of procedures, Hospital Episode Statistics—April, 1991, to March, 1995
184
THE LANCET • Vol 358 • July 21, 2001
For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
with increasing age seen in other centres (figure 2). The peak was apparent in the larger procedure groupings (interatrial repair, atrioventricular septal defect, ventricular septal defect; not shown) but other groups were too small to judge. Using 3-month age-groups, we estimated the total of excess deaths as 34·1 (slightly different from previous estimates due to the finer agestratification): those operated on in the first 3 months of life contributed 16·8 deaths and those in the final 3 months before their first birthday contributed 7·8—ie, about 25% of the total.
40
Proportion of operations (%)
35 30 25 20
Other centres
15 Bristol
10 5 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Age at operation (months)
Figure 2: Age at operation for open procedures by centre Hospital Episode Statistics: April, 1991, to March, 1995.
1·03–2·30) in epoch 2 to 1·95 (1·40–2·72) in epoch 3. When adjusted for procedure groupings, the odds ratios were 1·62 (0·96–2·72) and 1·90 (1·22–2·95) respectively. The 1991–95 results were highly significant, even allowing for multiple comparisons. Table 3 also shows results of the sensitivity analysis. Within epoch 2, exclusion of the three centres where comparison of the two data sources gave more than 20% difference in death rates increased the unadjusted odds ratio to 1·73 (1·01–2·98) and the adjusted estimate for procedure groupings to 1·86 (0·93–3·73). Exclusion of procedure groups 2, 3, and 5 increased the unadjusted odds ratio in epoch 2 to 2·38 (1·44–3·94) and the adjusted estimate for procedure groupings to 2·88 (95%I, 1·5–5·53). In epoch 3, odds ratios remained relatively unchanged after removal of the three centres and procedure groupings. Results with HES during epoch 3 were highly significant, and correction for deaths missed by the HES made little difference to the odds ratios. The final scenario in the sensitivity analysis, which selects the highest and lowest mortality within the two data sources for Bristol and non-Bristol centres, reduced unadjusted odds ratios to 1·83 (1·31–2·54) and the adjusted by-procedure-groupings estimate to 1·46 (0·97–2·21); only the former retained significance. Differences in case-mix between Bristol and other centres were not of a size or direction to account for the findings (table 4): surgeons at Bristol carried out fewer of their open operations on infants younger than 90 days than those elsewhere, and more on those aged between 90 days and 1 year; mortality in children with Down’s syndrome from centres other than Bristol was not significantly greater than children operated on without this disorder; Bristol had a much lower level of transferred patients, who have higher mortality, than elsewhere; Bristol admitted a smaller proportion of emergencies than other units. We also found that mortality did not01art vary/4468(2) significantly by quintile of Ref number socioeconomic deprivation in patients in other centres (6–7%). Editor Zoe Closer examination of age in months at operation showed aAuthor peak of activity in Bristol in month 11 in open procedures, by contrast with a steady decline in activity Illustrator
Milvia
Discussion We did a retrospective analysis of two national datasets to ascertain whether Bristol Royal Infirmary was an outlier with regard to mortality after paediatric cardiac surgery. The main finding was that Bristol was indeed an outlier, and not merely “bottom of the league”. There was an excess mortality between 1991 and March, 1995, for open operations in children younger than 1 year, and a mortality of around double that in other centres. The most compelling part of the data was the size of the discrepancy between the outcomes seen at Bristol and those seen elsewhere, especially for interatrial and other repairs to correct transposition of great arteries, and repair of complete atrial ventricular septal defects. Only one other centre showed evidence over the two data sources of divergent performance for open operations; this was at Harefield, for children older than 1 year. The surgical performance at Harefield—a national and international centre for transplant surgery—has been the subject of a recent report.13 Several possible explanations for the apparent excess mortality in Bristol need to be considered. We excluded chance as a likely explanation, since results for Bristol were highly significant, even when allowing multiple comparisons, and were robust up to extreme assumptions. However, issues of data quality that could potentially lead to bias must be addressed. If centres differ in the quality of their data, more meticulous centres might appear to have poorer results. Reasonable agreement between data sources improves the credibility of any conclusions.2 In particular, HES seemed to record 99% of 30-day postoperative deaths in hospital, although if deaths occurring outside hospital were included, the capture rate was 92%.11 Even if each data source were of perfect quality according to its own internal criteria, there would inevitably be disagreements on measures of activity and outcomes owing to the different data sources adopting different definitions. This problem is made worse by coding differences. Given that the data sources have such clear limitations, one could ask whether any reliable conclusions can be drawn. Statistical significance alone is not a sufficient guide, since it indicates only quantifiable random error and not systematic reporting or coding biases. The crucial issue is not whether the data are “correct”, since they manifestly contain errors, but whether such errors are likely to be great enough to mask the underlying pattern in the data. The sensitivity analyses give some indication of the extent of uncertainty of estimates of the excess mortality in Bristol. In our view, the size of the observed divergent performance in Bristol is such that reasonable variations in assumptions are not sufficient to Special instructions (PLEASE MARK WITH RED SPOT IF URGENT) cast doubt on our results. A further possible explanation for our findings might be differences in characteristics of children attending the different centres (case-mix). Surgeons at Bristol did a smaller proportion of their open operations on children
Journal
185
THE LANCET • Vol 358 • July 21, 2001
Section APPROVED c
m
y
k
Placed image
Repro purposes only
colour For personal use.mono Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
younger than 90 days compared with those elsewhere, yet mortality in Bristol was higher in all three age-groups. The number of operations peaked in infants aged 11 months, and the pattern for timing of surgery suggests that operations done just before an infant’s first birthday were delayed, rather than brought forward—a factor that accounts for about 25% of the excess deaths. None of the other variables considered could explain the high mortality at Bristol. What lessons can be learned more generally with regard to comparisons of clinical performance? Overall, there needs to be clarity as to the precise objectives of any comparative exercise. The crucial distinction is whether or not the aim is to carry out a-priori surveillance of health statistics to detect areas of concern about clinical performance. This type of investigation is analogous to cluster detection, for which there are many problems including sensitivity, specificity, and multiple significance testing.14 Alternatively, discrepant performance (of a unit or individual) might be investigated within a particular (preselected) clinical specialty. The emphasis of this study was estimation of excess mortality in a single centre, after concerns that were raised about paediatric cardiac surgical performance at Bristol Royal Infirmary. Our findings for Bristol indicate that data needed to be accumulated over many years to detect modest but important differences in mortality rates. Because of the limitations in existing data sources, only gross divergence could, in the past, have been identified with any degree of confidence. For example, if the mortality rate for open operations in children younger than 1 year at Bristol had been 50% higher than elsewhere (rather than 100% higher), the possibility that the difference had arisen through a combination of chance, data-quality issues, and case-mix could not have been excluded so readily. Other methods for assessing performance within a particular specialty have been proposed, including simple league tables15 and methods for statistical process control.16 Although the former corresponds to the first stage of our analysis, we allowed for uncertainty in rankings. Statistical process control is a simple graphical procedure intended to identify centres 3 SE from the mean, which corresponds to a two-sided p value of 0·003. This technique is similar to our third significance testing stage, although we adjusted for procedure groupings. Of course, none of these statistical approaches, if used for surveillance, addresses which area of clinical performance to investigate. This choice needs to be made on clinical grounds, otherwise issues of multiple significance testing remain. In the UK, there has been a particular focus on performance comparisons owing to recent high-profile cases including Bristol.1,17–19 If, as the UK government has stated, the national health service is to deliver high quality, cost-effective care that leads to improved health through guidance, audit, and best practice,21–23 it needs high quality information and linkage between administrative and clinical systems, mortality, and disease registers.24 Any future developments of routine data systems need to address the issue of how best to ensure that data are clinically valid and meaningful, based on a sense of ownership of the data by clinicians. Our methods are applicable more generally to the analysis of clinical and administrative data in other areas of clinical performance. We suggest that when comparing institutions, outcomes can and should be riskadjusted where feasible (ie, for mix of procedures), although this division might only be into broad groups
186
because too much stratification reduces precision. Statistical methods, as used here, can also prevent undue attention to spurious ranking into league tables.8 We recommend the informed introduction of appropriate statistical procedures for comparisons of institutional and individual clinical performance to provide better information for doctors, health-service managers, and patients. Contributors Paul Aylin, Stephen Evans, Gordon Murray, and David Spiegelhalter devised the original research, developed the protocols, and obtained funding. Paul Aylin, Bernadette Alves, Adrian Cook, Nicky Best, Stephen Evans, Audrey Lawrence, Gordon Murray, John Pollock, and David Spiegelhalter analysed data. Paul Aylin, David Spiegelhalter, and Paul Elliot were responsible for drafting and editing the paper, and all investigators contributed comments on the final draft.
Acknowledgments We thank Ruth Chadwick of the Bristol Royal Infirmary (BRI) Inquiry team, who commissioned this work on behalf of the Inquiry Panel; James Bennett, Alex Bottle, and Bruna Catena for assistance in producing reports; Leslie Hamilton and Kate Bull for advice on coding; Brian Hurwitz for commenting on the paper; and Alison Macfarlane, Klim McPherson, Mike Campbell, Robert Curnow, and Steve Gallivan of the BRI Inquiry’s Expert Group. The research on which this report is based was funded by the BRI Inquiry.
References 1 2
3
4
5
6
7
8
9 10
11
12 13
14
The Bristol Royal Infirmary Inquiry. www.bristol-inquiry.org.uk/ (accessed July 3, 2001). Spiegelhalter D, Evans SJW, Aylin P, Murray G. Overview of statistical evidence presented to the Bristol Royal Infirmary Inquiry concerning the nature and outcomes of paediatric cardiac surgical services at Bristol relative to other specialist centres from 1984 to 1995. www.bristol-inquiry.org.uk/Documents/statistical overview report.pdf (accessed July 3, 2001). The Inquiry’s approach to making use of relevant data sources. www.bristol-inquiry.org.uk/inqpro.htm#approach (accessed July 3, 2001). Evans SJW. A report on local data relating to children who received cardiac surgery under the terms of reference of the Bristol Royal Infirmary Inquiry. www.bristol-inquiry.org.uk/Documents/Local Data (Evans).pdf (accessed July 3, 2001). Murray G, Lawrence AE, Pollock J. A report on the UK Cardiac Surgical Register and the South West Congenital Heart Register for the Bristol Royal Infirmary Inquiry: a statistical analysis and review of the key data sources relevant to the Inquiry’s remit. www.bristolinquiry.org.uk/Documents/UKCSRSWCHR (Murray).pdf (accessed July 3, 2001). Aylin P, Alves B, Cook A, et al. Analysis of Hospital Episode Statistics for the Bristol Royal Infirmary Inquiry. www.bristolinquiry.org.uk/Documents/Hospital Episode Statistics (Aylin).pdf (accessed July 3, 2001). Inquiry Secretariat. Synthesis of statistical sources: a note on expert consultation on key analytical issues. www.bristolinquiry.org.uk/Documents/synthesis28.pdf (accessed July 3, 2001). Marshall EC, Spiegelhalter DJ. League tables of in vitro fertilisation clinics: how confident can we be about the rankings? BMJ 1998; 316: 1701–04. Gelman A, Carlin JB, Stern HS, Rubin D. Bayesian data analysis. London: Chapman and Hall, 1996. Spiegelhalter D. An initial synthesis of statistical sources concerning the nature and outcomes of paediatric cardiac surgical services at Bristol relative to other specialist centres from 1984 to 1995. www.bristol-inquiry.org.uk/Documents/synthesis2.pdf (accessed July 3, 2001). Murray G, Lawrence A, Boyd J. Linkage of Hospital Episode Statistics (HES) data to Office for National Statistics (ONS) mortality records. www.bristol-inquiry.org.uk/Documents/Hes ons linkage report.pdf (accessed July 3, 2001). Carstairs V, Morris R. Deprivation and health in Scotland. Aberdeen: Aberdeen University Press, 1991. Hunter S, Spiegelhalter D, Hamilton L. Review of clinical outcomes in children with congenital heart disease at Harefield hospital (1984–1999). www.rbh.nthames.nhs.uk/GENERAL/Press/ HHreport/report.htm (accessed July 5, 2001). Elliott P, Wakefield JC. Bias and confounding in spatial epidemiology. In: Elliott P, Wakefield JC, Best NG, Briggs DJ, eds.
THE LANCET • Vol 358 • July 21, 2001
For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
15
16
17
18
Spatial epidemiology: methods and applications. Oxford: Oxford University Press, 2000: 68–84. Department of Health. Quality and performance in the NHS. Performance indicators: July, 2000. www.doh.gov.uk/ nhsperformanceindicators/index.htm (accessed July 3, 2001). Mohammed MA, Cheng KK, Rouse A, Marshall T. Bristol, Shipman, and clinical governance: Shewhart’s forgotten lessons. Lancet 2001; 357: 463–67. The inquiry into the management of care of children receiving complex heart surgery at the Bristol Royal Infirmary. Interim report: removal and retention of human material. www.bristolinquiry.org.uk/interim/index.htm (accessed July 3, 2001). Horton R. The real lessons from Harold Frederick Shipman. Lancet 2001; 357: 82–83.
19 Ramsay S. Surgeon struck off in Canada, disgraced in UK. Lancet 2000; 356: 403. 20 Secretary of State for Health. The new NHS. London: Stationery Office, 1997 (Cm 3807). 21 National Institute for Clinical Excellence. www.nice.org.uk/nice-web/ (accessed July 3, 2001). 22 Commission for Health Improvement. www.chi.nhs.uk/ (accessed July 3, 2001). 23 NHS Executive. The NHS Performance Assessment Framework. www.doh.gov.uk/nhsexec/nhspaf.htm (accessed July 3, 2001). 24 Select Committee on Science and Technology. Human genetic databases: challenges and opportunities. London: Stationery Office, 2001 (HL Paper 57).
Clinical picture: Disseminated tuberculosis Bohdan Bybel, Randall C Starling, Richard C Brunken
During investigations for ischaemic cardiomyopathy, a 58-year-old middle-eastern man underwent 18-Fluoro-2-deoxy-glucose (FDG) positron emission tomography (PET) of the chest. Unexpectedly, multiple foci of intense lung and bony thorax uptake (figure, left) were noted. Diagnostic possibilities included hypermetabolic neoplastic lesions, active non-infectious granulomatous disease, or acute infection. Magnetic resonance imaging of the lumbar spine suggested osteomyelitis with abscess formation (figure, middle). The chest radiograph (figure, right) demonstrated right upper paravertebral and left retrocrural soft tissue masses (arrows). A computed tomography-guided aspiration of the paraspinous fluid collection showed Mycobacterium tuberculosis. Usually used to evaluate single pulmonary nodules and other suspected malignant diseases, FDG-PET clarifies equivocal findings. However, positive FDG-PET studies need to be interpreted with caution. Cleveland Clinic Foundation, 9500 Euclid Avenue, Cleveland, OH 44195 (B Bybel R Starling MD, R C Brunken MD)
THE LANCET • Vol 358 • July 21, 2001
FRCPS,
187
For personal use. Only reproduce with permission from The Lancet Publishing Group.
Comparison of UK paediatric cardiac surgical performance by analysis of routinely collected data 1984–96: was Bristol an outlier? Paul Aylin, Bernadette Alves, Nicky Best, Adrian Cook, Paul Elliott, Stephen J W Evans, Audrey E Lawrence, Gordon D Murray, John Pollock, David Spiegelhalter
Summary Background Reports of high mortality after paediatric cardiac surgery at the Bristol Royal Infirmary, UK, led to the establishment of an independent public inquiry. A key question was whether or not the mortality statistics in Bristol were unusual compared with other specialist centres. To answer this question, we did a retrospective analysis of mortality in the UK using two datasets. Methods Data from the UK Cardiac Surgical Register (CSR; January, 1984, to March, 1996) and Hospital Episode Statistics (HES; April, 1991, to December, 1995) were obtained for all 12 major centres in which paediatric cardiac surgery is done in the UK. The main outcome measure was mortality within 30 days of a cardiac surgical procedure. We estimated excess deaths in Bristol using a random-effects model derived from the remaining 11 centres. Additionally, a sensitivity analysis was done and case-mix examined. Findings For children younger than 1 year, in open operations, the mortality rate in Bristol was around double that of the other centres during 1991–95: within the CSR, there were 19·0 excess deaths (95% interval 2–32) among 43 deaths; and in HES, there were 24·1 excess deaths (12–34) among 41 deaths recorded. There was no strong evidence for excess mortality in Bristol for closed operations or for open operations in children older than 1 year. Interpretation Our results suggest that Bristol was an outlier, and we do not believe that statistical variation, systematic bias in data collection, case-mix, or data quality can explain a divergence in performance of this size. Lancet 2001; 358: 181–87
Division of Primary Care and Population Health Sciences, Imperial College School of Medicine, London, UK (P Aylin FFPHM, B Alves BA, N Best PhD, A Cook MSc, Prof P Elliott FRCP); Medicines Control Agency, London (Prof S J W Evans MSc); Lawrence Research, Bishopton (A E Lawrence MSc); Department of Community Health Sciences, University of Edinburgh Medical School, Edinburgh (Prof G D Murray PhD, J Pollock BSc); and MRC Biostatistics Unit, Institute of Public Health, Cambridge (D Spiegelhalter PhD) Correspondence to: Dr Paul Aylin, Department of Epidemiology and Public Health, Imperial College School of Medicine, St Mary’s Campus, Norfolk Place, London W2 1PG, UK (e-mail: [email protected])
THE LANCET • Vol 358 • July 21, 2001
Introduction Reports of high mortality within paediatric cardiac surgery units at the Bristol Royal Infirmary, UK, led to the establishment of an independent public inquiry.1 One of the key questions was whether or not the mortality statistics in Bristol were unusual compared with other specialist centres. This paper presents an analysis of data on paediatric cardiac surgery carried out at Bristol and 11 other major centres in England. We have previously shown that, within Bristol, levels of activity (numbers of cases and deaths) recorded in seven available datasets held for purposes of clinical audit or management (and, in one case, collated specifically for the Inquiry) were broadly similar, with no large discrepancies that could not be accounted for.2–6 Of these datasets, two were national: Hospital Episode Statistics (HES) and the UK Cardiac Surgical Register (CSR). Despite the limitations of routine data, national comparisons with these two sets of data could be undertaken. Although our study concerns outcome of paediatric cardiac surgery, its approach and methodology is applicable more generally to comparison of performance across a wide range of specialties.
Methods Data from the CSR (January, 1984, to March, 1996) and HES (April, 1991, to December, 1995) were obtained for all 12 major centres in England in which paediatric cardiac surgery is done. Two age-groups were defined: children younger than 1 year, and children aged between 1 and 15 years. Two different classifications of operative procedures were used: the first comprised broad classes of either open (requiring cardiopulmonary bypass) or closed operations, and the second consisted of 13 procedure groupings (11 open, two closed).7 Because the two data sources were available for different times within the study period, four epochs were defined:2 (1) January, 1984, to December, 1987; (2) January, 1988, to December, 1990; (3) January, 1991, to March, 1995; and (4) April, 1995, to December, 1995. For the HES data, mortality rate was based on admissions for whom discharge status was known (discharge home, transfer to another hospital, or death in hospital). Admissions with unknown discharge status were excluded. Our main outcome measure was mortality within 30 days of the procedure. Three stages of analysis were undertaken. First, overall mortality within each epoch was estimated with 95% CIs, and centres were ranked with 95% intervals for each rank.8 Second, expected mortality in Bristol was estimated, allowing for betweencentre variability. Specifically, within each procedure group we modelled between-centre variability using a variance-components (random-effect) model9 for the 11 centres (excluding Bristol), in which mortality rates (on a logit scale) were assumed drawn from a normal population distribution:10 the precisions of these normal
181
For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
distributions were further allowed to vary between procedure groups. The expected number of deaths in a hypothetical centre with Bristol’s level of activity was then predicted,6 and excess deaths in Bristol were estimated by subtracting these expected numbers from the observed numbers of deaths. The entire analysis was then repeated in turn for each of the remaining centres. The third stage comprised a traditional analysis to assess statistical significance and its robustness to alternative assumptions. A sensitivity analysis used four scenarios that contrasted mortality at Bristol with the average of the mortality rates in the other centres, by use of a logistic regression model to adjust for differences between centre and the mix of open surgical procedures. In the first scenario, we excluded from the analysis three centres for which comparison of the two data sources gave more than 20% difference in death rates; in the second, we removed procedure groups 2, 3, and 5 (interatrial and other repairs to correct transposition of
great arteries, and repair of complete atrial ventricular septal defects), which were prominent contributors to the observed overall excess mortality; in the third, we corrected undercount in the HES for all deaths within 30 days, estimated from a mortality linkage study,11 and recalculated the mortality rates for each centre; and in the fourth scenario, for each procedure group, we selected either HES or CSR data depending on which had the lower mortality for Bristol, and then, for the remaining centres, we chose the data source which had the higher mortality. Two-sided p values of less than 0·004 were sought as a Bonferroni adjustment for multiple comparisons. To examine possible confounding by case-mix, we obtained from HES: information on age at operation, proportions of children with a mention of Downs’ syndrome in any diagnosis field, the proportion of transfers from other units, and the proportions of cases admitted as an emergency, and compared mortality by
CSR data 1984–87, younger than 1 year
CSR data 1988–90, younger than 1 year
CSR data 1991–Mar 95, younger than 1 year
HES data 1991–Mar 95, younger than 1 year
Centre Mortality
Centre Mortality Approximate 95% interval
Centre Mortality Approximate 95% interval
Centre Mortality Approximate 95% interval
1 31/108 (29%) 2 22/107 (21%) 3 35/135 (26%) 4 14/45 (31%) 5 26/104 (25%) 6 34/198 (17%) 7 25/184 (14%) 8 57/362 (16%) 9 11/79 (14%) 10 34/90 (38%) 11 57/438 (13%) 12 21/121 (17%)
1 43/181 (24%) 2 27/200 (13%) 3 26/157 (17%) 4 15/142 (11%) 5 36/217 (17%) 6 49/417 (12%) 7 27/253 (11%) 8 57/369 (15%) 9 28/214 (13%) 10 31/184 (17%) 11 67/740 (9%) 12 32/268 (12%)
1 41/143 (29%) 2 25/187 (13%) 3 24/323 (7%) 4 23/122 (19%) 5 25/164 (15%) 6 42/405 (10%) 7 24/239 (10%) 8 53/482 (11%) 9 26/195 (13%) 10 25/177 (14%) 11 58/581 (10%) 12 31/301 (10%)
Total 367/1971 (19%)
Total 438/3342 (13%)
Total 397/3319 (12%)
CSR data 1988–90, 1–15 years
CSR data 1991–Mar 95, 1–15 years
HES data 1991–Mar 95, 1–15 years
Centre Mortality Approximate 95% interval
Centre Mortality Approximate 95% interval
Centre Mortality Approximate 95% interval
Centre Mortality Approximate 95% interval
1 24/284 (8%) 2 8/203 3 17/219 (8%) 4 0/15 (0%) 5 21/233 (9%) 6 49/567 (9%) 7 11/258 (4%) 8 52/459 (11%) 9 17/155 (11%) 10 32/213 (15%) 11 16/202 (8%) 12 14/380 (4%)
1 37/304 (12%) 2 10/132 (8%) 3 27/288 (9%) 4 3/107 (3%) 5 16/186 (9%) 6 44/397 (11%) 7 8/238 (3%) 8 37/776 (5%) 9 7/164 (4%) 10 28/161 (17%) 11 25/570 (4%) 12 19/300 (6%)
1 28/382 (7%) 2 6/343 (2%) 3 17/199 (9%) 4 9/67 (13%) 5 20/330 (6%) 6 33/623 (5%) 7 9/359 (3%) 8 18/525 (3%) 9 12/345 (3%) 10 48/373 (13%) 11 25/723 (3%) 12 35/616 (6%)
1 21/314 (7%) 2 5/216 (2%) 3 20/406 (5%) 4 6/168 (4%) 5 10/270 (4%) 6 27/738 (4%) 7 6/249 (2%) 8 32/642 (5%) 9 14/232 (6%) 10 35/292 (12%) 11 14/527 (3%) 12 25/471 (5%)
Total 261/3188 (8%)
Total 261/3623 (7%)
Total 260/4885 (5%)
Total 215/4525 (5%)
Approximate 95% interval
1 16/63 (25%) 2 11/66 (17%) 3 10/36 (28%) 4 0/0 (0%) 5 23/83 (28%) 6 48/242 (20%) 7 19/186 (10%) 8 55/236 (23%) 9 15/68 (22%) 10 28/109 (26%) 11 30/77 (39%) 12 28/187 (15%)
Total 283/1353 (21%)
CSR data 1984–87, 1–15 years
0 10 20 30 40
0 10 20 30 40
0 10 20 30 40
0 10 20 30 40
Surgical mortality rate (%)
Figure 1: Mortality from open procedures by centre derived from Hospital Episode Statistics (HES) and Cardiac Surgical Register (CSR) Centre 1=Bristol Royal Infirmary. HES mortality rates based on admissions with known outcome.
182
THE LANCET • Vol 358 • July 21, 2001
For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
Type of operation and data source
Epoch
All open operations CSR 1 (1984–87) 2 (1988–90) 3 (1991–95) 4 (1995–96)* Total (1984–96) HES 3 (1991–Mar 1995) 4 (1995)* Total (1991–95)
Children younger than 1 year
Children aged 1–15 years
Mortality in Bristol (deaths/cases)
Mortality elsewhere (deaths/cases)
Expected Excess deaths
Mortality in Bristol (deaths/cases)
Mortality elsewhere (deaths/cases)
Expected deaths
Excess
16/63 (25%) 31/108 (29%) 43/181 (24%) 3/50 (6%) 93/402 (23%) 41/143 (29%) 2/24 (8%) 43/167 (26%)
275/1308 (21%) 336/1863 (18%) 395/3161 (12%) 126/1049 (12%) 1132/7381 (15%) 356/3176 (11%) 68/563 (12%) 424/3739 (11%)
14·0 22·3 24·0 6·0 66·3 16·9 2·8 19·7
2·0 8·7 19·0 3·0 26·7 24·1 1·8 22·3
24/284 (8%) 37/304 (12%) 28/382 (7%) 2/136 (1%) 91/1106 (8%) 21/314 (7%) 0/87 21/401 (5%)
242/2989 (8%) 225/3333 (7%) 232/4508 (5%) 42/1305 (3%) 741/12 135 (6%) 194/4211 (5%) 31/695 (4%) 225/4906 (5%)
23·3 22·4 22·8 4·4 72·9 15·0 3·7 18·7
0·7 14·6 5·2 2·4 18·1 6·0 3·7 2·3
112/1851 (6%) 96/1750 (5%) 57/1839 (3%) 18/658 (3%) 283/6098 (5%) 78/1784 (4%) 25/357 (9%) 103/2141 (5%)
9·4 7·9 6·2 1·5 25·0 6·9 2·8 9·7
8·6 4·1 1·2 1·5 10·0 0·1 2·8 2·7
3/120 (2%) 4/127 (3%) 3/88 (3%) 1/24 (4%) 11/359 (3%) 0/89 1/28 (4%) 1/117 (1%)
21/1293 (2%) 21/1002 (2%) 21/792 (3%) 3/233 (1%) 66/3320 (2%) 15/893 (2%) 0/111 15/1004 (1%)
2·0 2·6 2·5 0·3 7·4 1·7 0 1·7
1·0 1·4 0·5 0·7 3·6 1·7 1·0 0·7
13·0 19·0 17·1 2·4 51·5 14·8 2·7 17·5
3·0 7·0 12·9 0·4 22·5 27·2 1·7 25·5
13·7 12·7 12·2 1·7 40·3 12·0 1·7 13·7
2·3 11·3 2·8 1·7 14·7 4·0 1·7 2·3
All closed operations CSR 1 (1984–87) 18/154 (12%) 2 (1988–90) 12/152 (8%) 3 (1991–95) 5/179 (3%) 4 (1995–96)* 0/54 Total (1984–96) 35/539 (6%) HES 3 (1991–Mar 1995) 7/153 (5%) 4 (1995)* 0/31 Total (1991–95) 7/184 (4%) 11 open procedure groups (adjusted for groupings) CSR 1 (1984–87) .. 2 (1988–90) .. 3 (1991–95) .. 4 (1995–96)* .. Total (1984–96) .. HES 3 (1991–Mar 1995) .. 4 (1995)* .. Total (1991–95) ..
.. .. .. .. .. .. .. ..
.. .. .. .. .. .. .. ..
.. .. .. .. .. .. .. ..
CSR=Cardiac Surgical Register. HES=Hospital Episode Statistics. HES mortality rates based on admissions with known outcomes. *Simplified analysis, excluding random effects.
Table 1: Observed and expected number of deaths and excess mortality in Bristol relative to other 11 centres in England
levels of socioeconomic deprivation (obtained from the Carstairs deprivation score12 for census enumeration districts linked to postcodes of residence). For open operations, the number and mortality rate at each month of age was calculated for each centre.
the third highest mortality rate (95% interval 1–6) reported to the CSR during 1988–90, and the highest rate in both the CSR (1–2) and HES (1–1) during 1991–95. Only one other centre, centre 10 (Harefield hospital), showed consistent evidence across the two sources of data of divergent performance, and this was for open operations in children older than 1 year. For closed procedures (not shown), no centres showed consistent evidence of divergent performance. Table 1 summarises mortality results for open and closed classes of procedures and the 11 open procedure
Results Figure 1 shows mortality by centre for open operations. Between-centre variability in outcomes reported to the CSR declined over the three epochs. In children younger than 1 year, out of the 12 centres, Bristol (centre 1) had Procedure
Mortality according to CSR Bristol
Open procedure groups G1 Tetralogy of Fallot 0/2 (0%) G2 Interatrial TGA 0/3 (0%)* G3 Other TGAs 10/36 (28%)* (switch) G4 Repair of TAPVD 6/18 (33%) G5 Repair of AVSD 8/32 (25%) (complete not partial) G6 Closure of secundum 2/5 (40%) and sinus venosus ASD G7 Closure of VSD 0/51 (0%) G8 Truncus arteriosus 2/7 (29%) G9 Fontan type 1/1 (100%) operations G10 Aortic, pulmonary valve, 1/1 (100%) and paravalve procedures G11 Mitral valve procedures 0/1 (0%) Closed procedure groups G12 Closed shunts G13 Coarctation procedures
0/30 (0%)
Elsewhere
Mortality according to HES ObExExcess served pected
Bristol
Elsewhere
Observed
Expected
Excess
22/270 (8%) 0 14/50 (28%)* 0 72/573 (13%)* 10
0·2 0·8 5
0·2 0·8 5
0/3 (0%) 2/15 (13%) 10/13 (77%)
18/281 (6%) 11/98 (11%) 60/573 (10%)
0 2 10
0·2 1·6 1·5
0·2 0·4 8·5
21/153 (14%) 44/331 (13%)
6 8
2·6 4·5
3·4 3·5
5/14 (36%) 11/23 (48%)
23/166 (14%) 48/402 (12%)
5 11
2 3
3 8·0
2/129 (2%)
2
0·1
1·9
5/10 (50%)
11/149 (7%)
5
0·7
4·3
18/689 (3%) 26/102 (25%) 7/21 (33%)
0 2 1
1·4 1·9 0·4
1·4 0·1 0·6
0/47 (0%) 3/4 (75%) 2/4 (50%)
45/811 (6%) 31/97 (32%) 20/118 (17%)
0 3 2
2·7 1·3 0·7
2·7 1·7 1·3
24/179 (13%)
1
0·2
0·8
2/4 (50%)
25/250 (10%)
2
0·4
1·6
5/36 (14%)
0
0·2
0·2
2/3 (67%)
12/51 (24%)
2
0·7
1·3
9/464 (2%)
0
0·6
0·6
3/38 (8%) 2/64 (3%)
58/598 (10%) 17/426 (4%)
3 2
3·8 2·6
0·8 0·6
CSR=Cardiac Surgical Register. HES=Hospital Episode Statistics. HES mortality rates based on admissions with known outcomes. ASD=atrial septal defect. AVSD=atrioventricular septal defect. TAPVD=total anomalous pulmonary-venous drainage. TGA=transposition of great arteries. VSD=ventricular septal defect. *Data are unreliable since seems to have been misuse of these groupings for operations such as “Mustard” and “Senning”.
Table 2: Observed and expected mortality relative to other 11 centres in England: 11 open and two closed procedure groups on children younger than 1 year during epoch 3 (April, 1991, to March, 1995)
THE LANCET • Vol 358 • July 21, 2001
183
For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
Data source and epoch
CSR 2 (1988–90)
3 (1991–95)
HES 3 (1991–95)
HES+CSR 3 (1991–95)
Analysis
Unadjusted for procedure grouping
Adjusted for procedure grouping
Odds ratio* (95% interval)
p
Odds ratio* (95% interval)
p
Baseline 1 2 Baseline 1 2
1·54 (1·03–2·30) 1·73 (1·01–2·98) 2·38 (1·44–3·94) 1·95 (1·40–2·72) 1.98 (1·27–3·10) 1·96 (1·28–3·01)
0·04 0·04 0·001 0·0001 0·002 0·002
1·62 (0·96–2·72) 1·86 (0·93–3·73) 2·88 (1·50–5·53) 1·90 (1·22–2·95) 1·97 (1·09–3·56) 1·90 (0·95–3·82)
0·07 0·08 0·001 0·004 0·03 0·07
Baseline 1 2 3
2·72 (1·92–3·86) 2·74 (1·71–4·38) 1·80 (1·10–2·95) 2·67 (1·71–4·17)
0·0001 0·0001 0·02 0·0001
3·43 (2·21–5·32) 3·46 (1·85–6·48) 2·58 (1·46–4·58) 3·20 (1·47–6·95)
0·0001 0·0001 0·001 0·003
4
1·83 (1·31–2·54)
0·0003
1·46 (0·97–2·21)
0·07
Analysis 1 removes three centres with highest discrepancies between Hospital Episode Statistics (HES) and Cardiac Surgical Register (CSR) data. Analysis 2 removes procedures for transposition of great arteries and atrioventricular septal defect. Analysis 3 corrects for undercount in HES statistics for all deaths within 30 days. Analysis 4 selects HES or CSR data for each procedure group dependent on which has the lower mortality (if Bristol), or the higher mortality (if elsewhere). HES mortality rates based on admissions with known outcomes. *Estimated ratio of odds of dying in Bristol compared with average odds of dying in other centres.
Table 3: Odds ratios and sensitivity analysis for open operations on children younger than 1 year
groupings, with data from the CSR and HES, for children younger or older than 1 year in Bristol and the 11 other centres. At ages younger than 1 year, in open operations within the CSR, there were 19·0 excess deaths (95% interval 2–32) among 43 deaths between January, 1991, and March, 1995—a time when the mortality rate in Bristol was around double that in other centres (24 vs 12%). There was no strong evidence for excess mortality for open operations in children older than 1 year during the same period. Reported mortality for open operations in children younger than 1 year fell in other centres from 21% in epoch 1 to 12% in epoch 3, whereas it did not improve at Bristol. There was no evidence of excess mortality in Bristol during epoch 4, although activity in Bristol was too small to draw any firm conclusion. Within the closed class of procedures, there was no strong evidence for excess mortality. HES data confirmed the pattern of deaths between April, 1991, and March, 1995: there were 24·1 excess deaths (12–34) among 41 deaths in Bristol and a mortality of 29% versus 11% elsewhere. There was no
strong evidence of excess mortality in closed procedures or in open operations in children older than 1 year. Within individual procedure groups for epoch 3 (1991–95) for children younger than 1 year, excess mortality with 95% confidence was identified within HES data for repairs to correct transposition of great arteries (other than interatrial), repair of complete atrial ventricular septal defects, and closure of secundum and sinus venosus atrial septal defects (table 2). After adjustment for procedure group, excess mortality at Bristol was retained, with 12·9 excess deaths in the CSR during 1991–95, and 27·2 in HES for the same epoch (table 1). Over all other centres, the corresponding maximum figure was 5·3 for CSR (centre 10) and 9·0 for HES (centre 4). Table 3 shows odds ratios and their 95% CI for the odds of dying in Bristol compared with the average odds of dying in the other centres for the 11 open procedure groupings (unadjusted and adjusted for procedure groupings) for children younger than 1 year. For the CSR, odds ratios increased from 1·54 (95% CI
Bristol
Elsewhere
Number of admissions
Mortality* (95%CI)
Number of admissions Mortality* (95%CI)
Age <90 days 90 days–1 year 1–15 years
37 (7%) 135 (27%) 333 (66%)
19/30 (63% [44–80]) 22/113 (19% [13–28]) 21/314 (7% [4–10])
1696 (22%) 1641 (21%) 4408 (57%)
254/1623 (16% [14–18]) 102/1562 (7% [5–8]) 195/4293 (5% [4–5])
Down’s syndrome Mentioned in any diagnosis field Not mentioned
52 (10%) 453 (90%)
7/51 (14% [6–26]) 55/406 (14% [10–17])
539 (7%) 7206 (93%)
43/537 (8% [6–10]) 508/6941 (7% [7–8])
Transfers From other units Non-transfers
32 (6%) 473 (94%)
15/26 (58% [37–77]) 47/431 (11% [8–14])
1707 (22%) 6038 (78%)
236/1656 (14% [12–16]) 315/5822 (5% [5–6])
Emergencies Emergency admissions Non-emergency admissions
36 (7%) 469 (93%)
13/28 (46% [28–66]) 49/429 (11% [9–15])
783 (10%) 6962 (90%)
88/751 (12% [10–14]) 463/6727 (7% [6–7])
Carstairs quintile* 1 (least deprived) 2 3 4 5 (most deprived) Unknown
95 (19%) 105 (21%) 134 (27%) 102 (20%) 67 (13%) 2 (0%)
9/82 (11% [5–20]) 10/88 (11% [6–20]) 11/122 (9% [5–16]) 11/88 (13% [6–21]) 1/55 (2% [0–10]) 1/1 (100% [3–100])
1161 (15%) 1274 (16%) 1281 (17%) 1327 (17%) 1634 (21%) 1068 (14%)
69/1116 (6% [5–8]) 86/1241 (7% [6–8]) 78/1253 (6% [5–8]) 91/1290 (7% [6–9]) 112/1592 (7% [6–8]) 97/977 (10% [8–12])
HES mortality rates based on admissions with known outcome. *Based on postcode of residence, where recorded.
Table 4: Status at admission and comorbidity for open class of procedures, Hospital Episode Statistics—April, 1991, to March, 1995
184
THE LANCET • Vol 358 • July 21, 2001
For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
with increasing age seen in other centres (figure 2). The peak was apparent in the larger procedure groupings (interatrial repair, atrioventricular septal defect, ventricular septal defect; not shown) but other groups were too small to judge. Using 3-month age-groups, we estimated the total of excess deaths as 34·1 (slightly different from previous estimates due to the finer agestratification): those operated on in the first 3 months of life contributed 16·8 deaths and those in the final 3 months before their first birthday contributed 7·8—ie, about 25% of the total.
40
Proportion of operations (%)
35 30 25 20
Other centres
15 Bristol
10 5 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Age at operation (months)
Figure 2: Age at operation for open procedures by centre Hospital Episode Statistics: April, 1991, to March, 1995.
1·03–2·30) in epoch 2 to 1·95 (1·40–2·72) in epoch 3. When adjusted for procedure groupings, the odds ratios were 1·62 (0·96–2·72) and 1·90 (1·22–2·95) respectively. The 1991–95 results were highly significant, even allowing for multiple comparisons. Table 3 also shows results of the sensitivity analysis. Within epoch 2, exclusion of the three centres where comparison of the two data sources gave more than 20% difference in death rates increased the unadjusted odds ratio to 1·73 (1·01–2·98) and the adjusted estimate for procedure groupings to 1·86 (0·93–3·73). Exclusion of procedure groups 2, 3, and 5 increased the unadjusted odds ratio in epoch 2 to 2·38 (1·44–3·94) and the adjusted estimate for procedure groupings to 2·88 (95%I, 1·5–5·53). In epoch 3, odds ratios remained relatively unchanged after removal of the three centres and procedure groupings. Results with HES during epoch 3 were highly significant, and correction for deaths missed by the HES made little difference to the odds ratios. The final scenario in the sensitivity analysis, which selects the highest and lowest mortality within the two data sources for Bristol and non-Bristol centres, reduced unadjusted odds ratios to 1·83 (1·31–2·54) and the adjusted by-procedure-groupings estimate to 1·46 (0·97–2·21); only the former retained significance. Differences in case-mix between Bristol and other centres were not of a size or direction to account for the findings (table 4): surgeons at Bristol carried out fewer of their open operations on infants younger than 90 days than those elsewhere, and more on those aged between 90 days and 1 year; mortality in children with Down’s syndrome from centres other than Bristol was not significantly greater than children operated on without this disorder; Bristol had a much lower level of transferred patients, who have higher mortality, than elsewhere; Bristol admitted a smaller proportion of emergencies than other units. We also found that mortality did not01art vary/4468(2) significantly by quintile of Ref number socioeconomic deprivation in patients in other centres (6–7%). Editor Zoe Closer examination of age in months at operation showed aAuthor peak of activity in Bristol in month 11 in open procedures, by contrast with a steady decline in activity Illustrator
Milvia
Discussion We did a retrospective analysis of two national datasets to ascertain whether Bristol Royal Infirmary was an outlier with regard to mortality after paediatric cardiac surgery. The main finding was that Bristol was indeed an outlier, and not merely “bottom of the league”. There was an excess mortality between 1991 and March, 1995, for open operations in children younger than 1 year, and a mortality of around double that in other centres. The most compelling part of the data was the size of the discrepancy between the outcomes seen at Bristol and those seen elsewhere, especially for interatrial and other repairs to correct transposition of great arteries, and repair of complete atrial ventricular septal defects. Only one other centre showed evidence over the two data sources of divergent performance for open operations; this was at Harefield, for children older than 1 year. The surgical performance at Harefield—a national and international centre for transplant surgery—has been the subject of a recent report.13 Several possible explanations for the apparent excess mortality in Bristol need to be considered. We excluded chance as a likely explanation, since results for Bristol were highly significant, even when allowing multiple comparisons, and were robust up to extreme assumptions. However, issues of data quality that could potentially lead to bias must be addressed. If centres differ in the quality of their data, more meticulous centres might appear to have poorer results. Reasonable agreement between data sources improves the credibility of any conclusions.2 In particular, HES seemed to record 99% of 30-day postoperative deaths in hospital, although if deaths occurring outside hospital were included, the capture rate was 92%.11 Even if each data source were of perfect quality according to its own internal criteria, there would inevitably be disagreements on measures of activity and outcomes owing to the different data sources adopting different definitions. This problem is made worse by coding differences. Given that the data sources have such clear limitations, one could ask whether any reliable conclusions can be drawn. Statistical significance alone is not a sufficient guide, since it indicates only quantifiable random error and not systematic reporting or coding biases. The crucial issue is not whether the data are “correct”, since they manifestly contain errors, but whether such errors are likely to be great enough to mask the underlying pattern in the data. The sensitivity analyses give some indication of the extent of uncertainty of estimates of the excess mortality in Bristol. In our view, the size of the observed divergent performance in Bristol is such that reasonable variations in assumptions are not sufficient to Special instructions (PLEASE MARK WITH RED SPOT IF URGENT) cast doubt on our results. A further possible explanation for our findings might be differences in characteristics of children attending the different centres (case-mix). Surgeons at Bristol did a smaller proportion of their open operations on children
Journal
185
THE LANCET • Vol 358 • July 21, 2001
Section APPROVED c
m
y
k
Placed image
Repro purposes only
colour For personal use.mono Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
younger than 90 days compared with those elsewhere, yet mortality in Bristol was higher in all three age-groups. The number of operations peaked in infants aged 11 months, and the pattern for timing of surgery suggests that operations done just before an infant’s first birthday were delayed, rather than brought forward—a factor that accounts for about 25% of the excess deaths. None of the other variables considered could explain the high mortality at Bristol. What lessons can be learned more generally with regard to comparisons of clinical performance? Overall, there needs to be clarity as to the precise objectives of any comparative exercise. The crucial distinction is whether or not the aim is to carry out a-priori surveillance of health statistics to detect areas of concern about clinical performance. This type of investigation is analogous to cluster detection, for which there are many problems including sensitivity, specificity, and multiple significance testing.14 Alternatively, discrepant performance (of a unit or individual) might be investigated within a particular (preselected) clinical specialty. The emphasis of this study was estimation of excess mortality in a single centre, after concerns that were raised about paediatric cardiac surgical performance at Bristol Royal Infirmary. Our findings for Bristol indicate that data needed to be accumulated over many years to detect modest but important differences in mortality rates. Because of the limitations in existing data sources, only gross divergence could, in the past, have been identified with any degree of confidence. For example, if the mortality rate for open operations in children younger than 1 year at Bristol had been 50% higher than elsewhere (rather than 100% higher), the possibility that the difference had arisen through a combination of chance, data-quality issues, and case-mix could not have been excluded so readily. Other methods for assessing performance within a particular specialty have been proposed, including simple league tables15 and methods for statistical process control.16 Although the former corresponds to the first stage of our analysis, we allowed for uncertainty in rankings. Statistical process control is a simple graphical procedure intended to identify centres 3 SE from the mean, which corresponds to a two-sided p value of 0·003. This technique is similar to our third significance testing stage, although we adjusted for procedure groupings. Of course, none of these statistical approaches, if used for surveillance, addresses which area of clinical performance to investigate. This choice needs to be made on clinical grounds, otherwise issues of multiple significance testing remain. In the UK, there has been a particular focus on performance comparisons owing to recent high-profile cases including Bristol.1,17–19 If, as the UK government has stated, the national health service is to deliver high quality, cost-effective care that leads to improved health through guidance, audit, and best practice,21–23 it needs high quality information and linkage between administrative and clinical systems, mortality, and disease registers.24 Any future developments of routine data systems need to address the issue of how best to ensure that data are clinically valid and meaningful, based on a sense of ownership of the data by clinicians. Our methods are applicable more generally to the analysis of clinical and administrative data in other areas of clinical performance. We suggest that when comparing institutions, outcomes can and should be riskadjusted where feasible (ie, for mix of procedures), although this division might only be into broad groups
186
because too much stratification reduces precision. Statistical methods, as used here, can also prevent undue attention to spurious ranking into league tables.8 We recommend the informed introduction of appropriate statistical procedures for comparisons of institutional and individual clinical performance to provide better information for doctors, health-service managers, and patients. Contributors Paul Aylin, Stephen Evans, Gordon Murray, and David Spiegelhalter devised the original research, developed the protocols, and obtained funding. Paul Aylin, Bernadette Alves, Adrian Cook, Nicky Best, Stephen Evans, Audrey Lawrence, Gordon Murray, John Pollock, and David Spiegelhalter analysed data. Paul Aylin, David Spiegelhalter, and Paul Elliot were responsible for drafting and editing the paper, and all investigators contributed comments on the final draft.
Acknowledgments We thank Ruth Chadwick of the Bristol Royal Infirmary (BRI) Inquiry team, who commissioned this work on behalf of the Inquiry Panel; James Bennett, Alex Bottle, and Bruna Catena for assistance in producing reports; Leslie Hamilton and Kate Bull for advice on coding; Brian Hurwitz for commenting on the paper; and Alison Macfarlane, Klim McPherson, Mike Campbell, Robert Curnow, and Steve Gallivan of the BRI Inquiry’s Expert Group. The research on which this report is based was funded by the BRI Inquiry.
References 1 2
3
4
5
6
7
8
9 10
11
12 13
14
The Bristol Royal Infirmary Inquiry. www.bristol-inquiry.org.uk/ (accessed July 3, 2001). Spiegelhalter D, Evans SJW, Aylin P, Murray G. Overview of statistical evidence presented to the Bristol Royal Infirmary Inquiry concerning the nature and outcomes of paediatric cardiac surgical services at Bristol relative to other specialist centres from 1984 to 1995. www.bristol-inquiry.org.uk/Documents/statistical overview report.pdf (accessed July 3, 2001). The Inquiry’s approach to making use of relevant data sources. www.bristol-inquiry.org.uk/inqpro.htm#approach (accessed July 3, 2001). Evans SJW. A report on local data relating to children who received cardiac surgery under the terms of reference of the Bristol Royal Infirmary Inquiry. www.bristol-inquiry.org.uk/Documents/Local Data (Evans).pdf (accessed July 3, 2001). Murray G, Lawrence AE, Pollock J. A report on the UK Cardiac Surgical Register and the South West Congenital Heart Register for the Bristol Royal Infirmary Inquiry: a statistical analysis and review of the key data sources relevant to the Inquiry’s remit. www.bristolinquiry.org.uk/Documents/UKCSRSWCHR (Murray).pdf (accessed July 3, 2001). Aylin P, Alves B, Cook A, et al. Analysis of Hospital Episode Statistics for the Bristol Royal Infirmary Inquiry. www.bristolinquiry.org.uk/Documents/Hospital Episode Statistics (Aylin).pdf (accessed July 3, 2001). Inquiry Secretariat. Synthesis of statistical sources: a note on expert consultation on key analytical issues. www.bristolinquiry.org.uk/Documents/synthesis28.pdf (accessed July 3, 2001). Marshall EC, Spiegelhalter DJ. League tables of in vitro fertilisation clinics: how confident can we be about the rankings? BMJ 1998; 316: 1701–04. Gelman A, Carlin JB, Stern HS, Rubin D. Bayesian data analysis. London: Chapman and Hall, 1996. Spiegelhalter D. An initial synthesis of statistical sources concerning the nature and outcomes of paediatric cardiac surgical services at Bristol relative to other specialist centres from 1984 to 1995. www.bristol-inquiry.org.uk/Documents/synthesis2.pdf (accessed July 3, 2001). Murray G, Lawrence A, Boyd J. Linkage of Hospital Episode Statistics (HES) data to Office for National Statistics (ONS) mortality records. www.bristol-inquiry.org.uk/Documents/Hes ons linkage report.pdf (accessed July 3, 2001). Carstairs V, Morris R. Deprivation and health in Scotland. Aberdeen: Aberdeen University Press, 1991. Hunter S, Spiegelhalter D, Hamilton L. Review of clinical outcomes in children with congenital heart disease at Harefield hospital (1984–1999). www.rbh.nthames.nhs.uk/GENERAL/Press/ HHreport/report.htm (accessed July 5, 2001). Elliott P, Wakefield JC. Bias and confounding in spatial epidemiology. In: Elliott P, Wakefield JC, Best NG, Briggs DJ, eds.
THE LANCET • Vol 358 • July 21, 2001
For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES
15
16
17
18
Spatial epidemiology: methods and applications. Oxford: Oxford University Press, 2000: 68–84. Department of Health. Quality and performance in the NHS. Performance indicators: July, 2000. www.doh.gov.uk/ nhsperformanceindicators/index.htm (accessed July 3, 2001). Mohammed MA, Cheng KK, Rouse A, Marshall T. Bristol, Shipman, and clinical governance: Shewhart’s forgotten lessons. Lancet 2001; 357: 463–67. The inquiry into the management of care of children receiving complex heart surgery at the Bristol Royal Infirmary. Interim report: removal and retention of human material. www.bristolinquiry.org.uk/interim/index.htm (accessed July 3, 2001). Horton R. The real lessons from Harold Frederick Shipman. Lancet 2001; 357: 82–83.
19 Ramsay S. Surgeon struck off in Canada, disgraced in UK. Lancet 2000; 356: 403. 20 Secretary of State for Health. The new NHS. London: Stationery Office, 1997 (Cm 3807). 21 National Institute for Clinical Excellence. www.nice.org.uk/nice-web/ (accessed July 3, 2001). 22 Commission for Health Improvement. www.chi.nhs.uk/ (accessed July 3, 2001). 23 NHS Executive. The NHS Performance Assessment Framework. www.doh.gov.uk/nhsexec/nhspaf.htm (accessed July 3, 2001). 24 Select Committee on Science and Technology. Human genetic databases: challenges and opportunities. London: Stationery Office, 2001 (HL Paper 57).
Clinical picture: Disseminated tuberculosis Bohdan Bybel, Randall C Starling, Richard C Brunken
During investigations for ischaemic cardiomyopathy, a 58-year-old middle-eastern man underwent 18-Fluoro-2-deoxy-glucose (FDG) positron emission tomography (PET) of the chest. Unexpectedly, multiple foci of intense lung and bony thorax uptake (figure, left) were noted. Diagnostic possibilities included hypermetabolic neoplastic lesions, active non-infectious granulomatous disease, or acute infection. Magnetic resonance imaging of the lumbar spine suggested osteomyelitis with abscess formation (figure, middle). The chest radiograph (figure, right) demonstrated right upper paravertebral and left retrocrural soft tissue masses (arrows). A computed tomography-guided aspiration of the paraspinous fluid collection showed Mycobacterium tuberculosis. Usually used to evaluate single pulmonary nodules and other suspected malignant diseases, FDG-PET clarifies equivocal findings. However, positive FDG-PET studies need to be interpreted with caution. Cleveland Clinic Foundation, 9500 Euclid Avenue, Cleveland, OH 44195 (B Bybel R Starling MD, R C Brunken MD)
THE LANCET • Vol 358 • July 21, 2001
FRCPS,
187
For personal use. Only reproduce with permission from The Lancet Publishing Group.