Operative mortality in colorectal cancer: prospective national studyBMJ 2003; 327 doi: https://doi.org/10.1136/bmj.327.7425.1196 (Published 20 November 2003) Cite this as: BMJ 2003;327:1196
- Appendix 1
The ACPGBI colorectal cancer model used a multilevel logistic regression analysis for identifying independent risk-factors of operative mortality and their interactions. Each risk factor was manually entered into the model starting from the most relevant; smallest P value, (such as age and co-morbidity) and adding each factor in-turn. By observing the odds ratios, the 95% confidence intervals for each new factor, and the change in the log-likelihood statistic we were able to ascertain whether each factor should remain the model or not.
Three levels of hierarchy were used in the ACPGBI model, placing the individual patient factors at the first level, named the "patient level". Individual hospitals and their annual case-volume were entered in the second level and UK regions were entered at the highest or third level. Conceptually the ACPGBI model can be viewed as a hierarchical system of regression equations with two distinct parts: (1) a "fixed part" which represents all the patient specific variables, and (2) a "random part" which exclusively models the variability of outcome between hospitals called the "hospital effect". The initial model estimates were derived using a second order penalised quasi-likelihood estimation. A Bayesian approach was then used, utilising diffuse priors (Gamma(ε ,ε ) prior distribution, where ε was set to 0.001) and the Gibbs re-sampling method with 50,000 iterations, to calculate confidence limits and to correct bias in the parameter estimation. The standard errors of the model estimates were therefore adjusted for the clustering of patients within hospitals and geographical regions.
The probability of operative mortality (log it Rijk) of patient i, in hospital j in region k was calculated using by the following equation:
log it Rijk = aijk + b1× (age)ijk + b2× (ASA)ijk + b3× (urgency)ijk + b4× (dukes)ijk
+ b5× (excision)ijk + b6× (ASA ´excision)ijk
aijk = a + vk + ujk
[vk] ~ N (0,Ωv) : Ωv = [σ2v]
[ujk] ~ N (0,Ωu) : Ωu = [σ2u]
where subscript i takes the value of 1 to the number of patients operated in a hospital j, in region k.
b1tob6 = fixed coefficientsor estimates for the patient risk factors and interaction term
a = constant for level 1 (fixed)
ujk = Level 2 random coefficient for hospital jk or "hospital-effect" which assumes a normal distribution with a mean of zero and variance σ2u
vk = Level 3 random coefficient for region k or "regional-effect" which assumes a normal distribution with a mean of zero and variance σ2v
The model can be utilised in the peri-operative setting for providing the best possible information to patients and their carers as part of the process of informed consent. It may also be used to compare outcomes between multidisciplinary colorectal cancer teams.
Appendix 2: List of hospitals that participated in the data collection for the ACPGBI study
Ashford & St Peter’s Hospitals, Chertsey; Bishop Auckland General Hospital; Borders General Hospital, Melrose; Broomfield Hospital, Chelmsford; Castle Hill Hospital, Hull; Central Middlesex Hospital, London; Charing Cross Hospital, London; Chelsea and Westminster Hospital; Colchester General Hospital, Colchester; Conquest Hospital, Hastings; Countess of Chester Hospital, Chester; Cumberland Infirmary, Carlisle; Darlington Memorial Hospital, Darlington; Derriford Hospital, Plymouth; Dorset County Hospital, Dorchester; Eastbourne District General Hospital, Eastbourne; Freeman Hospital, Newcastle Upon Tyne; Friarage Hospital, Northallerton; Glan Clywd Hospital, Bodelwyddan; Glenfield Hospital NHS Trust, Leicester; Good Hope Hospital, Sutton Coldfield; Hemel Hempstead General Hospital; Hereford County Hospital, Hereford; Hexham General Hospital; Homerton Hospital, London; James Cook University Hospital, Middlesbrough; Kettering General Hospital, Kettering; King’s College Hospital, London; Leeds General Infirmary, Leeds; Leicester General Hospital, Leicester; Leicester Royal Infirmary, Leicester; Luton and Dunstable Hospital; Manchester Royal Infirmary, Manchester; Morriston Hospital, Swansea; Ninewells Hospital, Dundee; North Devon District Hospital, Barnstaple; North Tees General Hospital, Stockton On Tees; North Tyneside General Hospital, North Shields; Northwick Park and St Mark’ s Hospitals, Harrow; Perth Royal Infirmary, Perth; Peterborough District Hospital, Peterborough; Poole Hospital NHS Trust, Poole; Princess of Wales Hospital, Bridgend; Princess Royal Hospital, Haywards Heath; Queen Alexandra Hospital, Portsmouth; Queen Elizabeth Hospital, Gateshead; Queen Elizabeth Hospital, Woolwich; Queen Elizabeth II Hospital, Welwyn Garden City; Queen’s Medical Centre, Nottingham; Royal Albert Edward Infirmary, Wigan; Royal Berkshire and Battle Hospitals, Reading; Royal Bournemouth Hospital, Bournemouth; Royal Free Hampstead NHS Trust; Royal Gwent Hospital, Newport; Royal Hallamshire Hospital, Sheffield; Royal Surrey County Hospital, Guildford; Royal Victoria Hospital, Belfast; Royal Victoria Infirmary, Newcastle upon Tyne; Singleton Hospital, Swansea; South Manchester University Hospital; South Tyneside District Hospital, South Shields; Sunderland Royal Hospital, Sunderland; Torbay Hospital, Torquay; University Hospital Lewisham, London; University Hospital of Hartlepool; University Hospital of North Durham, Durham; Wansbeck General Hospital, Ashington; West Cumberland Hospital, Whitehaven; West Wales General Hospital, Carmarthen; Wexham Park Hospital, Slough; Whipps Cross Hospital; Withybush General Hospital, Haverfordwest; Worthing Hospital, Worthing.
- This Week In The BMJ Published: 20 November 2003; BMJ 327 doi:10.1136/bmj.327.7425.0-b
- CorrectionWhy do children have chronic abdominal pain, and what happens to them when they grow up? Population based cohort studyPublished: 28 August 2003; BMJ 327 doi:10.1136/bmj.327.7413.500
- RESEARCHUse of administrative data or clinical databases as predictors of risk of death in hospital: comparison of modelsPublished: 23 April 2007; BMJ doi:10.1136/bmj.39168.496366.55
- ResearchUse of administrative data or clinical databases as predictors of risk of death in hospital: comparison of modelsPublished: 17 May 2007; BMJ 334 doi:10.1136/bmj.39168.496366.55
- Non-specific effects of measles, mumps, and rubella (MMR) vaccination in high income setting: population based cohort study in the NetherlandsBMJ August 30, 2017, 358 j3862; DOI: https://doi.org/10.1136/bmj.j3862
- Estimates of burden and consequences of infants born small for gestational age in low and middle income countries with INTERGROWTH-21st standard: analysis of CHERG datasetsBMJ August 17, 2017, 358 j3677; DOI: https://doi.org/10.1136/bmj.j3677
- Neurodevelopmental outcome at 2 years for preterm children born at 22 to 34 weeks’ gestation in France in 2011: EPIPAGE-2 cohort studyBMJ August 16, 2017, 358 j3448; DOI: https://doi.org/10.1136/bmj.j3448
- Childhood intelligence in relation to major causes of death in 68 year follow-up: prospective population studyBMJ June 28, 2017, 357 j2708; DOI: https://doi.org/10.1136/bmj.j2708
- Yemen’s cholera epidemic may have peaked, says WHOBMJ June 28, 2017, 357 j3131; DOI: https://doi.org/10.1136/bmj.j3131
- Acute large bowel obstruction secondary to stage 4 colonic carcinoma in an elderly man with severe aortic stenosis: a therapeutic challenge
- Colonic stenting: a practical update
- A population-based comparison of the survival of patients with colorectal cancer in England, Norway and Sweden between 1996 and 2004
- Thirty-day postoperative mortality after colorectal cancer surgery in England
- Hidden danger, obvious opportunity: error and risk in the management of cancer
- Use of administrative data or clinical databases as predictors of risk of death in hospital: comparison of models
- Gastroenterology services in the UK. The burden of disease, and the organisation and delivery of services for gastrointestinal and liver disorders: a review of the evidence