Web supplement for Endovascular or open repair for ruptured abdominal aortic aneurysm: 30-day outcomes from the IMPROVE trial IMPROVE trial investigators Containing: page Risk differences for mortality outcomes 2 Multiple imputations 2 Details of causal analyses 2 Details of economic analyses 2 Additional Results Figures (1) 3 Additional Results Tables (5) 4 References 9 1
Table A. Unadjusted risk differences for the primary and secondary mortality endpoints Outcome Endovascular Strategy Open repair Risk difference 95% confidence interval p-value N=316 N=297 30-day mortality 24-hour mortality In-hospital mortality 35.4% 37.4% -1.9% (-9.6 to 5.7) 0.620 21.5% 19.2% 2.3% (-4.0 to 8.7) 0.475 36.4% 38.4% -2.0% (-9.7 to 5.7) 0.611 Multiple imputation models Multiple imputation by chained equations (MICE) was used to handle missing data in baseline covariates (1). A set of imputation models was specified, one for each variable with missing values. Each variable was then regressed on all other variables, including completely recorded baseline and follow-up variables. The full list of variables considered in the MICE approach is shown in Table B, together with the number of missing values and the chained imputation model used for each variable. Causal Analysis A causal analysis in the ruptured AAA population only was conducted to address the question of what the effect of an endovascular strategy vs. Open repair in this population would be if everyone had adhered to the IMPROVE policy design (that is a CT scan plus EVAR if found anatomically suitable vs. Open repair). The causal analysis estimates the effect of the interventions in patients as randomised on the primary outcome (30-day mortality) in a complier population (2). This provides an unbiased estimate of the true treatment effect, subject to certain modelling assumptions (3), which would not be possible with a per-protocol analysis. Patients who were randomised to the endovascular strategy, found to be not anatomically suitable and underwent open repair were classified as having adhered with randomisation. Any other reasons for not receiving the allocated treatment were classified as non-adherence. Patients who had no operation (palliated) were excluded from the analysis under the assumption that their outcome would be the same no matter what group they were randomised to. Patients who had a converted operation (EVAR converted to Open) in the arm randomised to the EVAR strategy who were anatomically suitable were treated as having adhered. Patients who had a converted operation after randomisation to Open repair, were treated as having non-adherence. Economic Analysis The cost analysis reported inpatient costs within 30 days of randomisation. Individual resource use data for each primary hospital admission and re-admissions (including re-interventions) were prospectively recorded on the case report forms. For each rupture repair the following resource use items were recorded: number and type of stents, grafts and other consumables used for EVAR and Open Repair, time spent in the emergency room and endovascular suite or operating theatre (all in minutes), critical care, other specialist (e.g. stroke and coronary care) units and routine wards (all in days). For each day in critical care the number of organs supported was also recorded, which enabled each critical care bed-day to be assigned to the appropriate Healthcare Resource Group (4). 2
Missing resource data were addressed with multiple imputation, which assumes that the data were missing at random (5), that is conditional on baseline covariates (e.g. sex, Hardman Index), other resource use items (e.g. time in theatre), and health outcomes (e.g. survival, time to discharge). The unit costs of the stents and consumables used for rupture repair were taken from manufacturers list prices and published sources (Table D). Salary costs for rupture repair were calculated by combining staffing levels reported from a survey of 10 IMPROVE trial centres (www.imperial.ac.uk/medicine/improvetrial) with published staffing costs (6). The costs per critical care bed-day by Healthcare Resource Group, and for routine wards, were taken from the Payment by Results database (7). Total costs at 30 days were calculated by combining the resource use with unit costs, and reported in 2011-12 prices (in GBP). In the base case, incremental costs were reported as unadjusted mean differences between randomised arms, together with 95% confidence intervals. The following sensitivity analyses were undertaken: adjustment for baseline covariates and clustering with a multilevel model (8); a regression model that assumed total costs followed a Gamma distribution; alternative unit costs for the stents used for EVAR; different assumptions about staffing levels in the operating theatre; and alternative assumptions about resource use for patients with other (non-aneurysm related) discharge diagnoses. All analyses were according to intention-to-treat, with sub-group analyses conducted by age, sex and Hardman Index, and multiple imputation for addressing missing values. Figure A Sensitivity analysis: Incremental cost within the first 30 days post-randomisation (95% CI), under the base case and alternative assumptions [intention-to-treat analysis] 3
Table B Variables considered for multiple imputations Variable Number of missing values Imputation model Baseline variables Randomised group 0 None required (EVAR strategy vs. Open repair) Age 0 None required Sex 0 None required Admission systolic blood pressure 12 Linear regression Admission diastolic blood pressure 12 Linear regression Loss of consciousness during care episode 27 Logistic regression Admission haemoglobin 6 Linear regression Admission creatinine 13 Linear regression Acute myocardial ischaemia 52 Logistic regression Lowest recorded systolic blood pressure 46 Linear regression Lowest recorded diastolic blood pressure 53 Linear regression Hospital 0 None required Maximum aortic diameter* 77 Truncated regression (>0) Aneurysm neck diameter* 313 Truncated regression (>0) Aneurysm neck angle* 270 Truncated regression (>0) Aneurysm neck length* 270 Truncated regression (>0) Outcome variables Death within 24 hours 0 None required Death within 30 days 0 None required * conditional on ruptured AAA, symptomatic AAA or Incidental AAA in other final diagnosis. Core laboratory analyses for aneurysm neck diameter, angle and length not complete. 4
Table C 30-day mortality by randomised group, adjusted logistic regression with multiple imputation Variable Odds Ratio 95% CI p-value (Z-test) Randomised group (endovascular strategy versus open repair) 0.94 (0.67, 1.33) 0.732 Age* (per 5-year increase) 1.19 (1.04, 1.36) 0.014 Sex (male versus female) 0.72 (0.48, 1.09) 0.121 Hardman Index (per 1 unit increase) 1.59 (1.25, 2.01) <0.001 * Full effect of age on 30-day mortality confounded by discrete version of age ( 76, >76) included in the Hardman Index. 5
Table D: Unit costs ( GBP) Description Unit Endovascular strategy Open repair Source Medical devices and parts EVAR stent and parts Patient 5 700 Manufacturer list prices a Vascular graft (straight) Patient 623 Manufacturer list prices Vascular graft (bifurcated) Patient 901 Manufacturer list prices Consumables EVAR package Patient 600 Manufacturer list prices Mechanical retractor Patient 90 Manufacturer list prices Cell Salvage Patient 74 Davies et al 2006 (9) Surgical instrument set Patient 51 51 Manufacturer list prices Anaesthetics & other drugs Patient 41 184 British National Formulary b Contrast agent ml 0.10 0.10 IMPROVE centres b Blood unit 132 132 NHS Blood and Transplant (9) Platelets unit 205 205 NHS Blood and Transplant Fresh frozen plasma unit 25 25 NHS Blood and Transplant CT scan unit 105 105 Department of Health (7) Emergency room Minute 0.40 0.40 Dixon et al 2009 (10) Overheads Theatre Minute 2.65 2.65 IMPROVE centres c Staff d Surgeon (consultant) Minute 2.20 2.20 PSSRU (6) Surgeon (registrar) Minute 1.16 1.16 PSSRU Anaesthetist (consultant) Minute 2.20 2.20 PSSRU Anaesthetist (registrar) Minute 1.16 1.16 PSSRU Senior House Officer Minute 0.83 PSSRU ODA Minute 0.58 0.58 PSSRU Scrub Nurse Minute 0.72 0.72 PSSRU Runner Minute 0.58 0.58 PSSRU Radiologist (consultant) Minute 2.20 PSSRU Radiologist (registrar) Minute 1.16 PSSRU Radiographer Minute 0.58 PSSRU Radiologist Nurse Minute 0.72 PSSRU Critical care ITU/HDU 1 organ supported Bed-day 630 630 Department of Health (7) ITU/HDU 2 organs supported Bed-day 870 870 Department of Health ITU/HDU 3 organs supported Bed-day 1214 1214 Department of Health ITU/HDU 4 organs supported Bed-day 1410 1410 Department of Health ITU/HDU 5 organs supported Bed-day 1587 1587 Department of Health ITU/HDU 6 organs supported Bed-day 1759 1759 Department of Health ITU/HDU 7 organs supported Bed-day 2000 2000 Department of Health Other Coronary care unit Bed-day 436 436 Department of Health Stroke unit Bed-day 309 309 Department of Health Routine ward Bed-day 260 260 Department of Health a Average (range from 5400 to 6500) list price of the common EVAR stents and parts supplied to NHS Hospitals for ruptured AAA according to manufacturers list prices (Medtronic Endurant and Cook Medical Zenith Flex). b Local and general anaesthesia components were taken from one IMPROVE centre. c Unit costs obtained from IMPROVE centres. d Typical levels of staff use in theatre were recorded in 10 IMPROVE centres (www.imperial.ac.uk/medicine/improvetrial). 6
Table E Incremental [95% CI] costs ( GBP) within the first 30 days of randomisation; by subgroup Subgroup Incremental cost [95% CI] a Age Age 77 Age > 77-508 [-3079 to 2063] -2088 [-4652 to 476] Sex Male Female -1878 [-3899 to -143] 1486 [-2371 to 5343] Hardman=0-1939 [-5827 to 1950] Hardman Index Hardman=1-416 [-3131 to 2300] Hardman=2-2181 [-6173 to 1811] a Results are following multiple imputation. Estimates were obtained from a regression model which included a randomised arm by subgroup interaction term. 7
Table F Alternative cost assumptions for sensitivity analyses Base case Sensitivity analysis Baseline covariates Unadjusted analysis Adjusted for age, sex and Hardman Index Clustering Distribution of costs Theatre staff Single-level analysis Normal See footnote Multilevel analysis Gamma Varying these according to survey responses EVAR devices Manufacturer list price ( 5,700) Cost per case ranging from 4,000 to 10,000 Patients with non-raaa operation Included in the analysis Excluded from the analysis Resource use for patients with non-raaa operation Patients assumed to stay in routine ward for the whole hospital stay Patients assumed to stay in critical care for 50% of their hospital stay Canadian and Scottish centres Included in the analysis Excluded from the analysis Information about theatre staff was obtained from a survey of 10 centres participating in the trial. The base case considered that open repair was conducted with 2* anaesthestists, 2* vascular surgeons, 1 nurse, 2 other theatre staff and endovascular repair with 2* anaesthestists, 2* vascular surgeons, 1 nurse, 2 other theatre staff, 1 radiographer, 1 radiologist. *includes one training grade. 8
References 1. White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and guidance for practice. Stat Med. 2011;30(4):377-99. 2. Cuzick J, Edwards R, Segnan N. Adjusting for non-compliance and contamination in randomized clinical trials. Stat Med. 1997;16(9):1017-29. 3 Bellamy SL, Lin JY, Ten Have TR.An introduction to causal modeling in clinical trials. Clin Trials. 2007;4(1):58-73. 4. NHS Information Standard Board. Critical Care Minimum Data Set. Health and Social Care Information Centre; 2012. 5. Rubin DB. Multiple imputation for nonresponse in surveys. New York ; Chichester: Wiley; 1987. xxix, 258 p p. 6. Curtis L. Unit costs of health and social care. University of Kent, UK: Personal Social Services Research Unit; 2012. 7. Department of Health. National Schedule of Reference Costs for NHS Trusts, 2011-2012. London: Department of Health; 2012. 8. Grieve R, Nixon R, Thompson SG, Normand C. Using multilevel models for assessing the variability of multinational resource use and cost data. Health economics. 2005;14(2):185-96. 9. Davies L, Brown TJ, Haynes S, Payne K, Elliott RA, McCollum C. Cost-effectiveness of cell salvage and alternative methods of minimising perioperative allogeneic blood transfusion: a systematic review and economic model. Health Technol Assess. 2006;10(44):iii-iv, ix-x, 1-210. 10. Dixon S, Mason S, Knowles E, Colwell B, Wardrope J, Snooks H, et al. Is it cost effective to introduce paramedic practitioners for older people to the ambulance service? Results of a cluster randomised controlled trial. Emerg Med J. 2009;26(6):446-51. 9