- Email: [email protected]
Methodological design for economic evaluation in Public Access Defibrillation (PAD) trial
Trial Design
Methodological design for economic evaluation in Public Access Defibrillation (PAD) trial Graham Nichol, MD, George Arthur Wells, PhD, Karen Kuntz, ScD, David Feeny, PhD,1 Will Longstreth, MD, Brian Mahoney, MD, Clay Mann, PhD, Ray Lucas, MD, Mark Henry, MD, Ella Huszti, MSc, and Alice Birnbaum, MA Seattle, Wash
Objective Our objective is to describe the rationale and methods for the economic analysis of the PAD trial. The objective of this analysis is to assess whether automated external defibrillators (AEDs) use by lay responders is good value for money. Methods
Design. This economic evaluation is being conducted concurrently with a randomized trial of (a) control— training to recognize arrest, access 911, and administer cardiopulmonary resuscitation (CPR) while awaiting arrival of emergency medical services providers versus (b) intervention—training to recognize arrest, access 911, administer CPR, and use an AED while awaiting emergency medical services providers. Lay responders in either group were trained to deliver the study intervention. Population. Participating sites identified distinct units with a population of at least 250 people aged z50 years. Outcome. The primary economic outcome is the incremental cost-effectiveness ratio of intervention versus control.
Results Nine hundred ninety-three units including 1260 public and residential locations were randomized. There were 30 survivors in the intervention group and 15 in the control group ( P = .03). Sampling will identify program and health care costs. A societal perspective will be adopted. Incremental cost effectiveness will be estimated by using bootstrapping and decision analytic modeling. Conclusion The study will demonstrate whether defibrillation by lay responders improves outcomes at reasonable cost. If so, then the thousands of lives will be improved annually. If not, then limited resources can be invested in other interventions. Our methods also provide a framework for economic evaluations of other interventions for acute cardiovascular events. (Am Heart J 2005;150:202-8.)
Approximately 60% of deaths due to coronary heart disease are attributed to sudden cardiac arrest.1 Survival after the onset of out-of-hospital sudden cardiac arrest is correlated with the time from the onset of cardiac arrest to defibrillation,2-4 but has not improved during the last 20 years.5,6 PAD is a novel strategy for cardiac arrest that consists of training and equipping lay responders to defibrillate while awaiting arrival of emergency medical services From the Harborview Center for Prehospital Research and Training, Harborview Medical Center, University of Washington, Seattle, Wash. 1
David Feeny has a proprietary interest in Health Utilities Incorporated, the firm that distributes copyrighted Health Utilities Index questionnaires and related materials. This study was supported in part by contract N01-HC-95177 from the National Heart, Lung, and Blood Institute, Bethesda, Md, with additional support from the American Heart Association, Dallas, Tex; Guidant Corporation, Indianapolis, Ind; Medtronic, Inc, Minneapolis, Minn; Cardiac Science/Survivalink, Inc, Minneapolis, Minn; Medtronic
ERS, Redmond, Wash; Philips Medical Systems, Heartstream Operation, Seattle, Wash; and Laerdal Medical Corporation, Wappingers Falls, NY. Submitted June 29, 2004; accepted September 14, 2004. Reprint requests: Graham Nichol, MD, Harborview Center for Prehospital Research and Training, Harboview Medical Center, University of Washington Harborview Prehospital Research and Training Center, 325 Ninth Avenue, Box 359727, Seattle, WA 98104. E-mail: [email protected] 0002-8703/$ - see front matter n 2005, Mosby, Inc. All rights reserved. doi:10.1016/j.ahj.2004.09.034
providers. PAD is effective,7 but of unknown costs. To educate clinicians about how to assess the economics of interventions for acute cardiovascular events, we report here the economic methods used in the Public Access Defibrillation (PAD) trial.
Methods Objective The objective of this economic evaluation is to estimate the incremental cost effectiveness of training lay responders to recognize arrest, access 911, and administer cardiopulmonary resuscitation (CPR) while awaiting arrival of the community emergency medical services providers versus training lay responders to recognize arrest, access 911, administer CPR, and use an automated external defibrillator (AED) while waiting.
Overview of randomized trial The methods and results for the effectiveness trial were previously reported ( Figure 1).7,8 Participating sites were urban and suburban communities served by emergency medical services systems that provide advanced life support. Each site identified distinct units within their service area (eg, buildings, public areas). Lay responders in both control and intervention units were trained to (a) recognize cardiac arrest, (b) dial 911, and (c) perform CPR.
American Heart Journal Volume 150, Number 2
Nichol et al 203
will be converted to 2004 US dollars. Future costs and effects will be discounted at 3% per annum.10 The incremental cost-effectiveness ratio of intervention versus control will be evaluated by estimating (a) the costs of all training, retraining, AED implementation and maintenance, emergency medical services use, emergency department (ED), hospitalization, and postdischarge care; (b) the healthrelated quality of life associated with treatment by lay responders; (c) a truncated incremental cost-effectiveness ratio based on the study data; (d) a long-term incremental cost-effectiveness ratio based on a decision analytic model; and (e) the robustness of study results by using uncertainty and sensitivity analyses.
Figure 1
Outcomes Incremental cost-effectiveness ratio. The incremental costeffectiveness ratio of intervention versus control will be expressed as the additional cost of training and equipping lay responders to use an AED per additional unit of effectiveness (in quality-adjusted life years) compared with training and equipping lay responders for CPR alone.11
CostIntervention CostControl Quality¯ adjusted life years Intervention Quality¯ adjusted life yearsControl
Costing methodology PAD trial design.
Units randomized to intervention also received training in application of an AED. The intervention group was designed to place a trained AED—equipped lay responder at the side of an individual experiencing cardiac arrest within 3 minutes of identification of the event. The study population was individuals with out-of-hospital sudden cardiac arrest. The primary effectiveness analysis included cases with definite cardiac arrest. Included were those with any defibrillatory shock administered at the scene, or first rhythm recorded of ventricular fibrillation, ventricular tachycardia, or asystole. Excluded were those with obvious traumatic injury or aged younger than 8 years. There were more survivors to hospital discharge in the units assigned to have responders trained in CPR and use of AEDs (30 survivors among 128 arrests) than there were in the units assigned to have responders trained in CPR only (15 among 107, P = .03, relative risk 2.0, 95% CI 1.07-3.77).
Economic methods Overview. Cost and health-related quality of life data were collected concurrently with and beyond the duration of followup of the effectiveness trial. Upon completion of data collection, the primary economic analysis will adopt a societal perspective, including everyone affected by the intervention and all health outcomes and costs that flow from it, regardless of who experiences them,9 all intended effects (eg, survival to discharge) and unintended effects (eg, long-term care or complications of implantable defibrillator insertion). All costs
The perspective, type, method, and frequency of measurement and method of costing of each resource are summarized in Table I. The cost of each intervention will be calculated by (a) itemizing the resources used, (b) costing each of these resources, and (c) calculating the total cost by multiplying the amount of each resource by its unit cost and adding the result.
Resource use Training, retraining, AED implementation and maintenance, and emergency medical services resource use will be measured by using standardized questionnaires and relevant fiscal budgets. Relevant resources include instructor and student time, room rental, manikins, disposables, and so on, required for these activities at each site. Emergency medical services resource use will be estimated on a per-call basis from the annual budgets of participating emergency medical services systems. Health care resource use during hospitalization will be measured by obtaining itemized bills (ie, Uniform Bill-92s). Use of postdischarge resources will be identified by patient (or proxy if the patient is cognitively impaired)12 by using a standardized questionnaire.13,14 Respondents will be interviewed by telephone monthly for up to 3 months postdischarge and then quarterly until study end. Because there will be few survivors followed for N1 year, long-term resource use (eg, N3 months postdischarge) will be estimated by using Antiarrhythmic versus Implantable Defibrillator trial data.15
Valuation of costs Estimation of costs will adhere to current standards.10 All costs will be expressed on a per patient basis. The present value for devices (including all that are missing or replaced)
American Heart Journal August 2005
204 Nichol et al
Table I. Costing methodology Resource
Method of measurement
Training of responders
Number and type of resource
Implementation, including AED Police and fire
Number and type of resource
Perspective Site
Third-party payer
Method of valuation
Unit hours
Index training, retraining Index implementation, replacement Index arrest
Annual budget
Emergency medical services ED
Unit hours
Index arrest
Annual budget
Number and type of visit
Medicare data
Hospitalization after cardiac arrest Physician fees
Number and type of resourcey
Index arrest, subsequent visits Hospitalization
Number and type of visit
1m3
Type, dose and frequency Number and type of hematology, biochemistry, and microbiology tests Number and type of radiographic studies, invasive and noninvasive cardiac Number and duration of stay
1m3 1m3
Hospital-specific cost/charge ratio (http://www.cms.hhs.gov/) Medicare Resource-Based Relative-Value Scale50,51 Average wholesale price52 Medicare data
1m3
Medicare data
Any admission
RUGS (http://www.cms.hhs.gov/)
Number and duration of stay
Any admission
Current employment status of primary caregiver and number of days lost because of care for patient
1m3
Average daily charge for private room (http://metlife.com) Age-gender adjusted wages
Medications Laboratory tests
Procedures
Chronic or rehabilitation care Nursing home Societal
Frequency
Patient, caregiver time missed from work
Annual budget, wages Average wholesale price
yIncludes hospital room, nursing, medications, laboratory tests, and procedures.
will be calculated by assuming an 8-year device life span. To calculate training and retraining costs, the number of units of each resource used will be multiplied by their unit cost. To calculate costs of ED or hospital care, standard costs will be assigned based on Current Procedural Terminology codes or the relevant hospital’s annual Medicare cost-to-charge report (available at http://www.cms.hhs.gov/ ). The cost of professional fees and outpatient tests and procedures will be estimated by using current Medicare reimbursement levels. Nursing home costs will be estimated by multiplying the duration of stay (in days) by the average per diem for a private room in a nursing home (http://www.metlife.com).
Sampling frame
Generic health-related quality of life16,17 will be measured by trained interviewers administering the Health Utilities Index ( HUI ) Mark 2 and 3 systems with a standard 1-week recall quarterly to all patients who survive to discharge.18
We are collecting data prospectively to describe the cost of training and retraining, implementation, and emergency medical services for patients with presumed cardiac arrest. These will be identified and included in the analysis because costs are incurred by transporting these patients to hospital, regardless of whether they subsequently are found to have definite sudden cardiac arrest. For hospital costs, we will enroll all patients admitted with definite sudden cardiac arrest during the intervention period. Hospital costs will be incurred for patients with other than definite cardiac arrest, but no data will be collected to describe these patients. Instead we will assume that such patients are balanced equally between groups and not further characterize their costs. For postdischarge costs, we will enroll all patients discharged alive. Finally, we will identify and include costs of hospitalization for any lay responder who experiences an adverse event, if any.
Human subjects
Sample size
The University of Washington Institutional Review Board approved the study with an Exception from Informed Consent. Survivors will provide consent before collection of any data related to their health-related quality of life, resource use, or postdischarge status.
The primary purpose of the collection of cost and healthrelated quality of life data will be to increase the precision in their estimates in the economic analysis rather than to detect a statistically significant difference between control and intervention.
Health-related quality of life
American Heart Journal Volume 150, Number 2
For this trial, the available sample will include approximately 526 cases of presumed cardiac arrest: 260 in the intervention group and 266 in the control group. Eighty cases of definite cardiac arrest were admitted to hospital and available for identification of the costs of index hospitalization (50 in intervention group; 30 in control group.) Of these, 45 were discharged alive and available for identification of postdischarge costs and health-related quality of life (30 in intervention group; 15 in control group.) The anticipated precision in HUI scores will be F0.012.
Nichol et al 205
Figure 2
Planned analysis Overview. The primary analysis will consider the cost effectiveness of intervention versus control in public locations. Statistical analyses will be performed with S Plus.19 Imputation will be performed with standard statistical software.20,21 Decision analytic modeling will be performed with DATA.22 The truncated incremental cost effectiveness of intervention versus control will be estimated by applying bootstrapping to within-trial data to estimate the variability in the incremental cost-effectiveness ratio. Decision analytic modeling will be used to estimate the long-term incremental cost-effectiveness ratio of intervention versus control. Quality-adjusted life years will be calculated by multiplying the duration of each health state (ie, each follow-up period) by the patient-specific HUI3 score for that period. We are aware that previous economic evaluations of health interventions usually applied statistical tests to evaluate for differences in the mean (median) cost and effect of either intervention to determine whether the incremental cost effectiveness should be evaluated.23 Rather than such sequential testing, recent developments in economic methodology suggest that the emphasis should be on the estimation of the joint density of cost and effect differences and the quantification of uncertainty about the incremental cost-effectiveness ratio.24 Therefore, the incremental costs versus incremental effects will be summarized descriptively.
Estimation of hospitalization or postdischarge costs as incomplete longitudinal data There is ongoing controversy about how cost data should be analyzed.25,26 Cost data are collected longitudinally and are subject to dropout and truncation. Such missing values will be accounted for by using multiple imputation methods27,28 to minimize bias. Informative dropout will be accounted for as necessary.
Accounting for cases not participating in economic substudy Some cases may decline participation in the economic substudy. If these differ from participating cases, then the results may lack internal as well as external validity. A common approach to accounting for such unobserved data is to use multivariate analysis to describe observed costs as a function of covariates based on participating cases. Then costs are estimated for nonparticipating cases. However, this method underestimates uncertainty.29 Therefore, multiple imputation methods will be used.
Decision model of initial events after out-of-hospital sudden cardiac arrest.
Bootstrapping For the analysis of truncated costs and effects, bootstrapping will be used to assess the variability in the cost (effects) of either intervention by using the study data (ie, observed and imputed) to calculate confidence intervals.30
Decision analytic modeling The incremental cost effectiveness of intervention versus control will be assessed by using a decision model to project the costs and health outcomes beyond the observational time horizon of the study because adoption of a shorter time horizon biases the analysis against the intervention. This model will represent the probability and value of each outcome, including death before hospitalization, in hospital, or after discharge. Similar tree structures will be used to model intervention versus control. The postdischarge prognosis of survivors of sudden cardiac arrest will be modeled as a series of 1-month cycles in a Markov model.31,32 The probability of transition from one state to another will be derived from the events observed in the effectiveness trial combined with population life tables ( http:// www.cdc.gov/nchs) to project the outcomes beyond the trial to the lifetime horizon. The cost of each health state will be based on the costing methodology described above. The healthrelated quality of life associated with each health state will be based on the self-reported quality of life ( HUI3) of survivors. The model will be structured as 2 subtrees, representing the initial events after out-of-hospital cardiac arrest ( Figure 2) and events after insertion of an implantable cardioverter defibrillator ( Figure 3). The primary analysis will consider the actual rate of use of implantable defibrillators observed in the trial. Sensitivity analyses will consider the implantation of defibrillators in all neurologically intact (ie, cerebral performance category = 1) survivors.
Variability analysis The analysis will distinguish between uncertainty and variability. Uncertainty refers to variation in costs and effects due to sampling and measurement error. Variability refers to heterogeneity in costs and effects between groups of
American Heart Journal August 2005
206 Nichol et al
Figure 3
other interventions especially interventions for heart disease ( http://www.hsph.harvard.edu/cearegistry/ ).
Cost-effectiveness acceptability curves The incremental cost-effectiveness ratio of intervention versus control will be illustrated by using cost-effectiveness acceptability curves.37 These probability plots show the proportion of the observed incremental cost-effectiveness density that lies below a threshold ratio (k), which represents the monetary value of a unit of health gain. Decision makers can interpret the data in light of their threshold willingness to pay for the incremental health outcome.
Discussion
Decision model of events after insertion of implantable defibrillator.
patients who exhibit systematic differences in cost, effects, or both. For example, the incremental cost effectiveness of implantable defibrillators is conditional upon left ventricular ejection fraction.33 Within the trial, variability may reflect differences between patients with respect to risk for clinical outcomes or resource use, or differences between emergency medical services systems with respect to the type and availability of providers. Multivariate analyses will be used to estimate the incremental cost effectiveness of intervention versus control in different strata (eg, age, residential location of arrest).
Uncertainty analysis The impact of uncertainty in clinical and economic data will be assessed by using the following techniques.34 First, confidence intervals will be calculated for costs, effects, and incremental cost effectiveness by using Monte Carlo simulation.35,36 Second, detailed sensitivity analyses will illustrate the effect of uncertainty around the value of key variables, including future effects and costs, upon the incremental cost effectiveness of each intervention.34 These will include extensive sensitivity analyses to evaluate whether cost effectiveness is dependent on the persistence of any observed differences in survival or health-related quality of life beyond the duration of follow-up. Sensitivity analyses will also adopt the perspective of a third-party payer, including only costs paid by health insurance. Finally, the discount rate will be varied from 0% to 10%.
Comparison of cost-effectiveness estimates to those for other interventions The estimated incremental cost effectiveness of intervention versus control will be compared to published estimates for
This trial-based evaluation of the economics of training and equipping lay responders to recognize cardiac arrest, access 911, administer CPR, and promptly use an AED in the setting of sudden cardiac arrest will provide high-quality controlled data to inform patients, physicians, members of the public, and decision makers about whether to implement PAD. Economic evaluations of medical technologies assess the effectiveness and cost of the technologies so that physicians, policy makers, and the general public can decide which technologies offer sufficient value for the money. The cost effectiveness of health care interventions must be demonstrated definitively if claims of their public health benefit are to have scientific credibility.38 Because some experts question the merits of training and equipping lay responders to defibrillate,39 while others endorse such interventions,40 we believe that this study will provide critical information to inform public policy. In addition, the rigorous methodology used in this study can be used as a framework for evaluation of the economics of interventions for those with life-threatening arrhythmias or other acute cardiovascular events. Although previous analyses suggest that improvements that decrease the time to defibrillation are likely to be economically attractive,41-43 the present study offers several advantages over these. First, it will collect cost and health-related quality of life data from the same patients in the context of an ongoing randomized trial so as to minimize the effect of chance, bias, and confounding. In contrast, implementation studies in casinos44 and airports45 may have overestimated the efficacy of defibrillation because they lacked concurrent control data. Second, concurrent economic evaluation yields timely estimates of the economics of an intervention rather than delayed results based on an analysis initiated after demonstration of therapeutic effectiveness. Timeliness is important so that policy makers can promptly determine if an intervention improves clinical outcomes at comparatively little cost.46 Third, the economic evaluation uses a reliable and valid measure of health-related quality of life.16,17,47 Use
American Heart Journal Volume 150, Number 2
of this index is consistent with current standards for economic evaluation of health technologies.10 It was previously used to assess interventions for individuals with sudden cardiac arrest.13,48 Fourth, the economic evaluation will consider all present and future costs rather than just those of the program and hospital. Because the ongoing costs after implantable defibrillator insertion or long-term rehabilitation care are likely to be large, failure to include such future costs is likely to underestimate the incremental costs of defibrillation. Fifth, the statistical analysis of observed effects and costs will use multiple imputation to account for missing values.27,28,49 These data are collected longitudinally rather than at a single point in time and are subject to missing values because of inability to collect existing data or because the data do not exist due to dropout and truncation. Traditional methods of analysis of cost data do not account for dropout or truncation. Nonetheless, this study has some limitations. First, because of the large variety of settings studied in this trial, some settings are less represented than others. Therefore, we may have limited precision in the cost effectiveness of defibrillation in some settings. Second, the study will only collect cost and quality of life data on patients who consent to follow-up. To the extent that patients who decline consent are more or less sick than those who accede, these data are susceptible to bias. Third, the use of lay responders who are trained in CPR in the control group may bias the trial against the intervention in that the magnitude of training in the control group exceeds that typically found in standard care. Despite these limitations, this study is the most rigorous analysis of the economics of out-of-hospital interventions conducted to date. It provides a framework for future economic evaluations of interventions for patients with acute cardiovascular events by describing methods of measuring costs and healthrelated quality of life in this population. It also describes methods of analyses to determine whether such interventions are good value for money (ie, low incremental cost-effectiveness ratio). This study will demonstrate whether defibrillation by lay responders improves outcomes at reasonable cost. If lay defibrillation is effective and inexpensive, then the thousands of lives can be improved annually. If not, then limited health care resources can be invested in other interventions that are good value for money. We acknowledge the constructive comments provided by 4 anonymous reviewers. This paper is dedicated to the memories of Thomas P. Holohan and Peter Frank who died September 11, 2001, in the World Trade Center.
Nichol et al 207
References 1. National Center for Health Statistics. Detailed diagnoses and procedures, National Hospital Discharge Survey, 1990. Hyattsville, Md; 1992. 2. Becker LB, Ostrander MP, Barrett J, et al. Outcomes of CPR in a large metropolitan area—where are the survivors? Ann Emerg Med 1991;355 - 61. 3. Lombardi G, Gallagher J, Gennis P. Outcome of out-of-hospital cardiac arrest in New York city. The Pre-Hospital Arrest Survival Evaluation (PHASE) study. JAMA 1994;678 - 83. 4. Holmberg M, Holmberg S, Herlitz J. An alternate estimate of the disappearance rate of ventricular fibrillation in out-of-hospital cardiac arrest in Sweden. Resuscitation 2001;219 - 20. 5. Cobb LA, Fahrenbruch CE, Olsufka M, et al. Changing incidence of out-of-hospital ventricular fibrillation, 1980-2000. JAMA 2002;3008 - 13. 6. Herlitz J, Bang A, Gunnarsson J, et al. Factors associated with survival to hospital discharge among patients hospitalised alive after out of hospital cardiac arrest: change in outcome over 20 years in the community of Goteborg, Sweden. Heart 2003;25 - 30. 7. Hallstrom AP, Ornato JP, Weisfeldt M, et al. Public-access defibrillation and survival after out-of-hospital cardiac arrest. N Engl J Med 2004;637 - 46. 8. PAD Investigators. The public access defibrillation: study design and rationale. Resuscitation 2003;135 - 47. 9. Russell LB, Siegel JE, Daniels N, et al. In: Weinstien MC, editor. Cost-effectiveness analysis as a guide to resource allocation in health: roles and limitations. Cost-effectiveness in health and medicine. Oxford University Press: New York; 1996. p. 3 - 24. 10. Gold MR, Siegel JE, Russell LB, et al. Appendix A: summary recommendations. Cost-effectiveness in health and medicine. Oxford University Press: New York; 1996. p. 425. 11. Drummond MF, O’Brien BJ, Stoddart GL, et al. Methods for the economic evaluation of health care programs. 2nd ed. Oxford University Press: Oxford; 1997. 12. Roccaforte WH, Burke WJ, Bayer BL, et al. Validation of a telephone version of the mini-mental state examination. J Am Geriatr Soc 1992;697 - 702. 13. Stiell IG, Wells GA, Spaite DW, et al. The Ontario Prehospital Advanced Life Support (OPALS) study: rationale and methodology for cardiac arrest patients. Ann Emerg Med 1998;180 - 90. 14. Stiell IG, Wells GA, Spaite DW, et al. The Ontario Prehospital Advanced Life Support (OPALS) study Part II: rationale and methodology for trauma and respiratory distress patients. OPALS study group. Ann Emerg Med 1999;256 - 62. 15. The Antiarrhythmics Versus Implantable Defibrillators (AVID) Investigators. A comparison of antiarrhythmic drug therapy with implantable defibrillators in patients resuscitated from near-fatal ventricular arrhythmias. N Engl J Med 1997;1576 - 83. 16. Torrance GW, Furlong W, Feeny D, et al. Multi-attribute preference functions: health utilities index. PharmacoEconomics 1995;503 - 20. 17. Feeny D, Furlong W, Boyle W, et al. Multi-attribute health status classification systems: health utilities index. PharmacoEconomics 1995;490 - 502. 18. Feeny D, Furlong W, Torrance GW, et al. Multiattribute and singleattribute utility functions for the health utilities index mark 3 system. Med Care 2002;113 - 28. 19. Anonymous. S-Plus 2000. Data Analysis Products Division, MathSoft Inc:Seattle (WA); 1999.
208 Nichol et al
20. Anonymous. SOLAS for missing data analysis. 3.0 ed. Statistical Solutions Inc. Boston (MA); 2000. 21. Smith DM. OSWALD. 3.4 ed. Lancaster University Department of Mathematics and Statistics: Lancaster; 2000. 22. TreeAge Software I. Decision Analysis by TreeAge. DATA for Windows. Version 3.5 User’s Manual. Boston, MA; 1999. 23. Barber JA, Thompson SG. Analysis and interpretation of cost data in randomised controlled trials: review of published studies. BMJ 1998;1195 - 2000. 24. Briggs AH, O’Brien BJ. The death of cost-minimization analysis? Health Econ Lett 2000;3 - 10. 25. Etzioni RD, Feuer EJ, Sullivan SD, et al. On the use of survival analysis techniques to estimate medical care costs. J Health Econ 1999;365 - 80. 26. Hallstrom AP, Sullivan SD. On estimating costs for economic evaluation in failure time studies. Med Care 1998;433 - 6. 27. Little RJA, Rubin DB. Statistical analysis with missing data. In: Watson GS, editor. Wiley series in probability and mathematical statistics. John Wiley and Sons: New York; 1987. 28. Rubin DB. Multiple imputation for nonresponse in surveys. Wiley: New York; 1987. 29. Graham JW, Schafer JL. On the performance of multiple imputation for multivariate data with small sample size. In: Hoyle RH, editor. Statistical strategies for small sample research. SAGE Publications: London; 1999. p. 1 - 32. 30. Efron B, Tibshirani RJ. An introduction to the bootstrap. Chapman and Hall: New York; 1993. 31. Sonnenberg FA, Beck JR. Markov models in medical decision making: a practical guide. Med Decis Making1993;322 - 38. 32. Naimark D, Krahn MD, Naglie G, et al. Primer on medical decision analysis: part 5-working with Markov processes. Med Decis Making 1997;152 - 9. 33. O’Brien BJ, Connolly SJ, Goeree R, et al. Cost-effectiveness of the implantable cardioverter-defibrillator: results from the Canadian Implantable Defibrillator Study (CIDS). Circulation 2001;1416 - 21. 34. Briggs A, Sculpher M, Buxton M. Uncertainty in the economic evaluation of health care technologies: the role of sensitivity analysis. Health Econ 1994;95 - 104. 35. Doubilet P, Begg CB, Weinstein MC, et al. Probabilistic sensitivity analysis using Monte Carlo simulation. A practical approach. Med Decis Making 1985;157 - 77. 36. Critchfield GC, Willard KE. Probabilistic analysis of decision trees using Monte Carlo simulation. Med Decis Making 1986;86 - 92.
American Heart Journal August 2005
37. van Hout BA, al Maiwenn J, Gordon GS, et al. Costs, effects and C/E ratios alongside a clinical trial. Health Econ 1994;309 - 19. 38. Dowie J. Clinical trials and economic evaluations? No, there are only evaluations. Health Econ 1997;87 - 9. 39. Pell JP, Sirel JM, Marsden AK, et al. Potential impact of public access defibrillators on survival after out of hospital cardiopulmonary arrest: retrospective cohort study. BMJ 2002;515. 40. Anonymous. Guidelines 2000 for cardiopulmonary resuscitation and emergency cardiovascular care. International consensus on science. Circulation 2000;1 - 291. 41. Nichol G, Hallstrom A, Ornato JP, et al. Potential cost-effectiveness of public access defibrillation in the United States. Circulation 1998;1315 - 20. 42. Nichol G, Laupacis A, Stiell I, et al. A cost-effectiveness analysis of potential improvements to emergency medical services for victims of out-of-hospital cardiac arrest. Ann Emerg Med 1996;711 - 20. 43. Groeneveld PW, Kwong JL, Liu Y, et al. Cost-effectiveness of automated external defibrillators on airlines. JAMA 2001;1482 - 9. 44. Valenzuela TD, Roe DJ, Nichol G, et al. Outcomes of rapid defibrillation by security officers after cardiac arrest in casinos. N Engl J Med 2000;1206 - 9. 45. Caffrey SL, Willoughby PJ, Pepe PE, et al. Public use of automated external defibrillators. N Engl J Med 2002;1242 - 7. 46. Bloom BS, Fendrick AM. Timing and timeliness in medical care evaluation. PharmacoEconomics 1996;183 - 7. 47. Boyle M, Furlong W, Torrance G, et al. Reliability of the health utilities index—Mark III used in the 1991 cycle 6 Canadian General Social Survey Health Questionnaire. Qual Life Res 1995;249 - 57. 48. Nichol G, Stiell IG, Hebert P, et al. What is the quality of life of survivors of cardiac arrest? A prospective study. Acad Emerg Med 1999;95 - 102. 49. Rubin DB, Schenker N. Multiple imputation for interval estimation from simple random samples with ignorable nonresponse. J Am Stat Assoc 1986;366 - 74. 50. Hsiao WC, Braun P, Dunn D, et al. Resource-based relative values: an overview. JAMA 1988;2347 - 53. 51. Hsiao WC, Braun P, Dunn D, et al. Results and policy implications of the resource based relative-value study. N Engl J Med 1988;881 - 8. 52. Anonymous. 2001 drug topics red bookMedical Economics Data Inc: New York; 2001. p. 800.
Methodological design for economic evaluation in Public Access Defibrillation (PAD) trial Graham Nichol, MD, George Arthur Wells, PhD, Karen Kuntz, ScD, David Feeny, PhD,1 Will Longstreth, MD, Brian Mahoney, MD, Clay Mann, PhD, Ray Lucas, MD, Mark Henry, MD, Ella Huszti, MSc, and Alice Birnbaum, MA Seattle, Wash
Objective Our objective is to describe the rationale and methods for the economic analysis of the PAD trial. The objective of this analysis is to assess whether automated external defibrillators (AEDs) use by lay responders is good value for money. Methods
Design. This economic evaluation is being conducted concurrently with a randomized trial of (a) control— training to recognize arrest, access 911, and administer cardiopulmonary resuscitation (CPR) while awaiting arrival of emergency medical services providers versus (b) intervention—training to recognize arrest, access 911, administer CPR, and use an AED while awaiting emergency medical services providers. Lay responders in either group were trained to deliver the study intervention. Population. Participating sites identified distinct units with a population of at least 250 people aged z50 years. Outcome. The primary economic outcome is the incremental cost-effectiveness ratio of intervention versus control.
Results Nine hundred ninety-three units including 1260 public and residential locations were randomized. There were 30 survivors in the intervention group and 15 in the control group ( P = .03). Sampling will identify program and health care costs. A societal perspective will be adopted. Incremental cost effectiveness will be estimated by using bootstrapping and decision analytic modeling. Conclusion The study will demonstrate whether defibrillation by lay responders improves outcomes at reasonable cost. If so, then the thousands of lives will be improved annually. If not, then limited resources can be invested in other interventions. Our methods also provide a framework for economic evaluations of other interventions for acute cardiovascular events. (Am Heart J 2005;150:202-8.)
Approximately 60% of deaths due to coronary heart disease are attributed to sudden cardiac arrest.1 Survival after the onset of out-of-hospital sudden cardiac arrest is correlated with the time from the onset of cardiac arrest to defibrillation,2-4 but has not improved during the last 20 years.5,6 PAD is a novel strategy for cardiac arrest that consists of training and equipping lay responders to defibrillate while awaiting arrival of emergency medical services From the Harborview Center for Prehospital Research and Training, Harborview Medical Center, University of Washington, Seattle, Wash. 1
David Feeny has a proprietary interest in Health Utilities Incorporated, the firm that distributes copyrighted Health Utilities Index questionnaires and related materials. This study was supported in part by contract N01-HC-95177 from the National Heart, Lung, and Blood Institute, Bethesda, Md, with additional support from the American Heart Association, Dallas, Tex; Guidant Corporation, Indianapolis, Ind; Medtronic, Inc, Minneapolis, Minn; Cardiac Science/Survivalink, Inc, Minneapolis, Minn; Medtronic
ERS, Redmond, Wash; Philips Medical Systems, Heartstream Operation, Seattle, Wash; and Laerdal Medical Corporation, Wappingers Falls, NY. Submitted June 29, 2004; accepted September 14, 2004. Reprint requests: Graham Nichol, MD, Harborview Center for Prehospital Research and Training, Harboview Medical Center, University of Washington Harborview Prehospital Research and Training Center, 325 Ninth Avenue, Box 359727, Seattle, WA 98104. E-mail: [email protected] 0002-8703/$ - see front matter n 2005, Mosby, Inc. All rights reserved. doi:10.1016/j.ahj.2004.09.034
providers. PAD is effective,7 but of unknown costs. To educate clinicians about how to assess the economics of interventions for acute cardiovascular events, we report here the economic methods used in the Public Access Defibrillation (PAD) trial.
Methods Objective The objective of this economic evaluation is to estimate the incremental cost effectiveness of training lay responders to recognize arrest, access 911, and administer cardiopulmonary resuscitation (CPR) while awaiting arrival of the community emergency medical services providers versus training lay responders to recognize arrest, access 911, administer CPR, and use an automated external defibrillator (AED) while waiting.
Overview of randomized trial The methods and results for the effectiveness trial were previously reported ( Figure 1).7,8 Participating sites were urban and suburban communities served by emergency medical services systems that provide advanced life support. Each site identified distinct units within their service area (eg, buildings, public areas). Lay responders in both control and intervention units were trained to (a) recognize cardiac arrest, (b) dial 911, and (c) perform CPR.
American Heart Journal Volume 150, Number 2
Nichol et al 203
will be converted to 2004 US dollars. Future costs and effects will be discounted at 3% per annum.10 The incremental cost-effectiveness ratio of intervention versus control will be evaluated by estimating (a) the costs of all training, retraining, AED implementation and maintenance, emergency medical services use, emergency department (ED), hospitalization, and postdischarge care; (b) the healthrelated quality of life associated with treatment by lay responders; (c) a truncated incremental cost-effectiveness ratio based on the study data; (d) a long-term incremental cost-effectiveness ratio based on a decision analytic model; and (e) the robustness of study results by using uncertainty and sensitivity analyses.
Figure 1
Outcomes Incremental cost-effectiveness ratio. The incremental costeffectiveness ratio of intervention versus control will be expressed as the additional cost of training and equipping lay responders to use an AED per additional unit of effectiveness (in quality-adjusted life years) compared with training and equipping lay responders for CPR alone.11
CostIntervention CostControl Quality¯ adjusted life years Intervention Quality¯ adjusted life yearsControl
Costing methodology PAD trial design.
Units randomized to intervention also received training in application of an AED. The intervention group was designed to place a trained AED—equipped lay responder at the side of an individual experiencing cardiac arrest within 3 minutes of identification of the event. The study population was individuals with out-of-hospital sudden cardiac arrest. The primary effectiveness analysis included cases with definite cardiac arrest. Included were those with any defibrillatory shock administered at the scene, or first rhythm recorded of ventricular fibrillation, ventricular tachycardia, or asystole. Excluded were those with obvious traumatic injury or aged younger than 8 years. There were more survivors to hospital discharge in the units assigned to have responders trained in CPR and use of AEDs (30 survivors among 128 arrests) than there were in the units assigned to have responders trained in CPR only (15 among 107, P = .03, relative risk 2.0, 95% CI 1.07-3.77).
Economic methods Overview. Cost and health-related quality of life data were collected concurrently with and beyond the duration of followup of the effectiveness trial. Upon completion of data collection, the primary economic analysis will adopt a societal perspective, including everyone affected by the intervention and all health outcomes and costs that flow from it, regardless of who experiences them,9 all intended effects (eg, survival to discharge) and unintended effects (eg, long-term care or complications of implantable defibrillator insertion). All costs
The perspective, type, method, and frequency of measurement and method of costing of each resource are summarized in Table I. The cost of each intervention will be calculated by (a) itemizing the resources used, (b) costing each of these resources, and (c) calculating the total cost by multiplying the amount of each resource by its unit cost and adding the result.
Resource use Training, retraining, AED implementation and maintenance, and emergency medical services resource use will be measured by using standardized questionnaires and relevant fiscal budgets. Relevant resources include instructor and student time, room rental, manikins, disposables, and so on, required for these activities at each site. Emergency medical services resource use will be estimated on a per-call basis from the annual budgets of participating emergency medical services systems. Health care resource use during hospitalization will be measured by obtaining itemized bills (ie, Uniform Bill-92s). Use of postdischarge resources will be identified by patient (or proxy if the patient is cognitively impaired)12 by using a standardized questionnaire.13,14 Respondents will be interviewed by telephone monthly for up to 3 months postdischarge and then quarterly until study end. Because there will be few survivors followed for N1 year, long-term resource use (eg, N3 months postdischarge) will be estimated by using Antiarrhythmic versus Implantable Defibrillator trial data.15
Valuation of costs Estimation of costs will adhere to current standards.10 All costs will be expressed on a per patient basis. The present value for devices (including all that are missing or replaced)
American Heart Journal August 2005
204 Nichol et al
Table I. Costing methodology Resource
Method of measurement
Training of responders
Number and type of resource
Implementation, including AED Police and fire
Number and type of resource
Perspective Site
Third-party payer
Method of valuation
Unit hours
Index training, retraining Index implementation, replacement Index arrest
Annual budget
Emergency medical services ED
Unit hours
Index arrest
Annual budget
Number and type of visit
Medicare data
Hospitalization after cardiac arrest Physician fees
Number and type of resourcey
Index arrest, subsequent visits Hospitalization
Number and type of visit
1m3
Type, dose and frequency Number and type of hematology, biochemistry, and microbiology tests Number and type of radiographic studies, invasive and noninvasive cardiac Number and duration of stay
1m3 1m3
Hospital-specific cost/charge ratio (http://www.cms.hhs.gov/) Medicare Resource-Based Relative-Value Scale50,51 Average wholesale price52 Medicare data
1m3
Medicare data
Any admission
RUGS (http://www.cms.hhs.gov/)
Number and duration of stay
Any admission
Current employment status of primary caregiver and number of days lost because of care for patient
1m3
Average daily charge for private room (http://metlife.com) Age-gender adjusted wages
Medications Laboratory tests
Procedures
Chronic or rehabilitation care Nursing home Societal
Frequency
Patient, caregiver time missed from work
Annual budget, wages Average wholesale price
yIncludes hospital room, nursing, medications, laboratory tests, and procedures.
will be calculated by assuming an 8-year device life span. To calculate training and retraining costs, the number of units of each resource used will be multiplied by their unit cost. To calculate costs of ED or hospital care, standard costs will be assigned based on Current Procedural Terminology codes or the relevant hospital’s annual Medicare cost-to-charge report (available at http://www.cms.hhs.gov/ ). The cost of professional fees and outpatient tests and procedures will be estimated by using current Medicare reimbursement levels. Nursing home costs will be estimated by multiplying the duration of stay (in days) by the average per diem for a private room in a nursing home (http://www.metlife.com).
Sampling frame
Generic health-related quality of life16,17 will be measured by trained interviewers administering the Health Utilities Index ( HUI ) Mark 2 and 3 systems with a standard 1-week recall quarterly to all patients who survive to discharge.18
We are collecting data prospectively to describe the cost of training and retraining, implementation, and emergency medical services for patients with presumed cardiac arrest. These will be identified and included in the analysis because costs are incurred by transporting these patients to hospital, regardless of whether they subsequently are found to have definite sudden cardiac arrest. For hospital costs, we will enroll all patients admitted with definite sudden cardiac arrest during the intervention period. Hospital costs will be incurred for patients with other than definite cardiac arrest, but no data will be collected to describe these patients. Instead we will assume that such patients are balanced equally between groups and not further characterize their costs. For postdischarge costs, we will enroll all patients discharged alive. Finally, we will identify and include costs of hospitalization for any lay responder who experiences an adverse event, if any.
Human subjects
Sample size
The University of Washington Institutional Review Board approved the study with an Exception from Informed Consent. Survivors will provide consent before collection of any data related to their health-related quality of life, resource use, or postdischarge status.
The primary purpose of the collection of cost and healthrelated quality of life data will be to increase the precision in their estimates in the economic analysis rather than to detect a statistically significant difference between control and intervention.
Health-related quality of life
American Heart Journal Volume 150, Number 2
For this trial, the available sample will include approximately 526 cases of presumed cardiac arrest: 260 in the intervention group and 266 in the control group. Eighty cases of definite cardiac arrest were admitted to hospital and available for identification of the costs of index hospitalization (50 in intervention group; 30 in control group.) Of these, 45 were discharged alive and available for identification of postdischarge costs and health-related quality of life (30 in intervention group; 15 in control group.) The anticipated precision in HUI scores will be F0.012.
Nichol et al 205
Figure 2
Planned analysis Overview. The primary analysis will consider the cost effectiveness of intervention versus control in public locations. Statistical analyses will be performed with S Plus.19 Imputation will be performed with standard statistical software.20,21 Decision analytic modeling will be performed with DATA.22 The truncated incremental cost effectiveness of intervention versus control will be estimated by applying bootstrapping to within-trial data to estimate the variability in the incremental cost-effectiveness ratio. Decision analytic modeling will be used to estimate the long-term incremental cost-effectiveness ratio of intervention versus control. Quality-adjusted life years will be calculated by multiplying the duration of each health state (ie, each follow-up period) by the patient-specific HUI3 score for that period. We are aware that previous economic evaluations of health interventions usually applied statistical tests to evaluate for differences in the mean (median) cost and effect of either intervention to determine whether the incremental cost effectiveness should be evaluated.23 Rather than such sequential testing, recent developments in economic methodology suggest that the emphasis should be on the estimation of the joint density of cost and effect differences and the quantification of uncertainty about the incremental cost-effectiveness ratio.24 Therefore, the incremental costs versus incremental effects will be summarized descriptively.
Estimation of hospitalization or postdischarge costs as incomplete longitudinal data There is ongoing controversy about how cost data should be analyzed.25,26 Cost data are collected longitudinally and are subject to dropout and truncation. Such missing values will be accounted for by using multiple imputation methods27,28 to minimize bias. Informative dropout will be accounted for as necessary.
Accounting for cases not participating in economic substudy Some cases may decline participation in the economic substudy. If these differ from participating cases, then the results may lack internal as well as external validity. A common approach to accounting for such unobserved data is to use multivariate analysis to describe observed costs as a function of covariates based on participating cases. Then costs are estimated for nonparticipating cases. However, this method underestimates uncertainty.29 Therefore, multiple imputation methods will be used.
Decision model of initial events after out-of-hospital sudden cardiac arrest.
Bootstrapping For the analysis of truncated costs and effects, bootstrapping will be used to assess the variability in the cost (effects) of either intervention by using the study data (ie, observed and imputed) to calculate confidence intervals.30
Decision analytic modeling The incremental cost effectiveness of intervention versus control will be assessed by using a decision model to project the costs and health outcomes beyond the observational time horizon of the study because adoption of a shorter time horizon biases the analysis against the intervention. This model will represent the probability and value of each outcome, including death before hospitalization, in hospital, or after discharge. Similar tree structures will be used to model intervention versus control. The postdischarge prognosis of survivors of sudden cardiac arrest will be modeled as a series of 1-month cycles in a Markov model.31,32 The probability of transition from one state to another will be derived from the events observed in the effectiveness trial combined with population life tables ( http:// www.cdc.gov/nchs) to project the outcomes beyond the trial to the lifetime horizon. The cost of each health state will be based on the costing methodology described above. The healthrelated quality of life associated with each health state will be based on the self-reported quality of life ( HUI3) of survivors. The model will be structured as 2 subtrees, representing the initial events after out-of-hospital cardiac arrest ( Figure 2) and events after insertion of an implantable cardioverter defibrillator ( Figure 3). The primary analysis will consider the actual rate of use of implantable defibrillators observed in the trial. Sensitivity analyses will consider the implantation of defibrillators in all neurologically intact (ie, cerebral performance category = 1) survivors.
Variability analysis The analysis will distinguish between uncertainty and variability. Uncertainty refers to variation in costs and effects due to sampling and measurement error. Variability refers to heterogeneity in costs and effects between groups of
American Heart Journal August 2005
206 Nichol et al
Figure 3
other interventions especially interventions for heart disease ( http://www.hsph.harvard.edu/cearegistry/ ).
Cost-effectiveness acceptability curves The incremental cost-effectiveness ratio of intervention versus control will be illustrated by using cost-effectiveness acceptability curves.37 These probability plots show the proportion of the observed incremental cost-effectiveness density that lies below a threshold ratio (k), which represents the monetary value of a unit of health gain. Decision makers can interpret the data in light of their threshold willingness to pay for the incremental health outcome.
Discussion
Decision model of events after insertion of implantable defibrillator.
patients who exhibit systematic differences in cost, effects, or both. For example, the incremental cost effectiveness of implantable defibrillators is conditional upon left ventricular ejection fraction.33 Within the trial, variability may reflect differences between patients with respect to risk for clinical outcomes or resource use, or differences between emergency medical services systems with respect to the type and availability of providers. Multivariate analyses will be used to estimate the incremental cost effectiveness of intervention versus control in different strata (eg, age, residential location of arrest).
Uncertainty analysis The impact of uncertainty in clinical and economic data will be assessed by using the following techniques.34 First, confidence intervals will be calculated for costs, effects, and incremental cost effectiveness by using Monte Carlo simulation.35,36 Second, detailed sensitivity analyses will illustrate the effect of uncertainty around the value of key variables, including future effects and costs, upon the incremental cost effectiveness of each intervention.34 These will include extensive sensitivity analyses to evaluate whether cost effectiveness is dependent on the persistence of any observed differences in survival or health-related quality of life beyond the duration of follow-up. Sensitivity analyses will also adopt the perspective of a third-party payer, including only costs paid by health insurance. Finally, the discount rate will be varied from 0% to 10%.
Comparison of cost-effectiveness estimates to those for other interventions The estimated incremental cost effectiveness of intervention versus control will be compared to published estimates for
This trial-based evaluation of the economics of training and equipping lay responders to recognize cardiac arrest, access 911, administer CPR, and promptly use an AED in the setting of sudden cardiac arrest will provide high-quality controlled data to inform patients, physicians, members of the public, and decision makers about whether to implement PAD. Economic evaluations of medical technologies assess the effectiveness and cost of the technologies so that physicians, policy makers, and the general public can decide which technologies offer sufficient value for the money. The cost effectiveness of health care interventions must be demonstrated definitively if claims of their public health benefit are to have scientific credibility.38 Because some experts question the merits of training and equipping lay responders to defibrillate,39 while others endorse such interventions,40 we believe that this study will provide critical information to inform public policy. In addition, the rigorous methodology used in this study can be used as a framework for evaluation of the economics of interventions for those with life-threatening arrhythmias or other acute cardiovascular events. Although previous analyses suggest that improvements that decrease the time to defibrillation are likely to be economically attractive,41-43 the present study offers several advantages over these. First, it will collect cost and health-related quality of life data from the same patients in the context of an ongoing randomized trial so as to minimize the effect of chance, bias, and confounding. In contrast, implementation studies in casinos44 and airports45 may have overestimated the efficacy of defibrillation because they lacked concurrent control data. Second, concurrent economic evaluation yields timely estimates of the economics of an intervention rather than delayed results based on an analysis initiated after demonstration of therapeutic effectiveness. Timeliness is important so that policy makers can promptly determine if an intervention improves clinical outcomes at comparatively little cost.46 Third, the economic evaluation uses a reliable and valid measure of health-related quality of life.16,17,47 Use
American Heart Journal Volume 150, Number 2
of this index is consistent with current standards for economic evaluation of health technologies.10 It was previously used to assess interventions for individuals with sudden cardiac arrest.13,48 Fourth, the economic evaluation will consider all present and future costs rather than just those of the program and hospital. Because the ongoing costs after implantable defibrillator insertion or long-term rehabilitation care are likely to be large, failure to include such future costs is likely to underestimate the incremental costs of defibrillation. Fifth, the statistical analysis of observed effects and costs will use multiple imputation to account for missing values.27,28,49 These data are collected longitudinally rather than at a single point in time and are subject to missing values because of inability to collect existing data or because the data do not exist due to dropout and truncation. Traditional methods of analysis of cost data do not account for dropout or truncation. Nonetheless, this study has some limitations. First, because of the large variety of settings studied in this trial, some settings are less represented than others. Therefore, we may have limited precision in the cost effectiveness of defibrillation in some settings. Second, the study will only collect cost and quality of life data on patients who consent to follow-up. To the extent that patients who decline consent are more or less sick than those who accede, these data are susceptible to bias. Third, the use of lay responders who are trained in CPR in the control group may bias the trial against the intervention in that the magnitude of training in the control group exceeds that typically found in standard care. Despite these limitations, this study is the most rigorous analysis of the economics of out-of-hospital interventions conducted to date. It provides a framework for future economic evaluations of interventions for patients with acute cardiovascular events by describing methods of measuring costs and healthrelated quality of life in this population. It also describes methods of analyses to determine whether such interventions are good value for money (ie, low incremental cost-effectiveness ratio). This study will demonstrate whether defibrillation by lay responders improves outcomes at reasonable cost. If lay defibrillation is effective and inexpensive, then the thousands of lives can be improved annually. If not, then limited health care resources can be invested in other interventions that are good value for money. We acknowledge the constructive comments provided by 4 anonymous reviewers. This paper is dedicated to the memories of Thomas P. Holohan and Peter Frank who died September 11, 2001, in the World Trade Center.
Nichol et al 207
References 1. National Center for Health Statistics. Detailed diagnoses and procedures, National Hospital Discharge Survey, 1990. Hyattsville, Md; 1992. 2. Becker LB, Ostrander MP, Barrett J, et al. Outcomes of CPR in a large metropolitan area—where are the survivors? Ann Emerg Med 1991;355 - 61. 3. Lombardi G, Gallagher J, Gennis P. Outcome of out-of-hospital cardiac arrest in New York city. The Pre-Hospital Arrest Survival Evaluation (PHASE) study. JAMA 1994;678 - 83. 4. Holmberg M, Holmberg S, Herlitz J. An alternate estimate of the disappearance rate of ventricular fibrillation in out-of-hospital cardiac arrest in Sweden. Resuscitation 2001;219 - 20. 5. Cobb LA, Fahrenbruch CE, Olsufka M, et al. Changing incidence of out-of-hospital ventricular fibrillation, 1980-2000. JAMA 2002;3008 - 13. 6. Herlitz J, Bang A, Gunnarsson J, et al. Factors associated with survival to hospital discharge among patients hospitalised alive after out of hospital cardiac arrest: change in outcome over 20 years in the community of Goteborg, Sweden. Heart 2003;25 - 30. 7. Hallstrom AP, Ornato JP, Weisfeldt M, et al. Public-access defibrillation and survival after out-of-hospital cardiac arrest. N Engl J Med 2004;637 - 46. 8. PAD Investigators. The public access defibrillation: study design and rationale. Resuscitation 2003;135 - 47. 9. Russell LB, Siegel JE, Daniels N, et al. In: Weinstien MC, editor. Cost-effectiveness analysis as a guide to resource allocation in health: roles and limitations. Cost-effectiveness in health and medicine. Oxford University Press: New York; 1996. p. 3 - 24. 10. Gold MR, Siegel JE, Russell LB, et al. Appendix A: summary recommendations. Cost-effectiveness in health and medicine. Oxford University Press: New York; 1996. p. 425. 11. Drummond MF, O’Brien BJ, Stoddart GL, et al. Methods for the economic evaluation of health care programs. 2nd ed. Oxford University Press: Oxford; 1997. 12. Roccaforte WH, Burke WJ, Bayer BL, et al. Validation of a telephone version of the mini-mental state examination. J Am Geriatr Soc 1992;697 - 702. 13. Stiell IG, Wells GA, Spaite DW, et al. The Ontario Prehospital Advanced Life Support (OPALS) study: rationale and methodology for cardiac arrest patients. Ann Emerg Med 1998;180 - 90. 14. Stiell IG, Wells GA, Spaite DW, et al. The Ontario Prehospital Advanced Life Support (OPALS) study Part II: rationale and methodology for trauma and respiratory distress patients. OPALS study group. Ann Emerg Med 1999;256 - 62. 15. The Antiarrhythmics Versus Implantable Defibrillators (AVID) Investigators. A comparison of antiarrhythmic drug therapy with implantable defibrillators in patients resuscitated from near-fatal ventricular arrhythmias. N Engl J Med 1997;1576 - 83. 16. Torrance GW, Furlong W, Feeny D, et al. Multi-attribute preference functions: health utilities index. PharmacoEconomics 1995;503 - 20. 17. Feeny D, Furlong W, Boyle W, et al. Multi-attribute health status classification systems: health utilities index. PharmacoEconomics 1995;490 - 502. 18. Feeny D, Furlong W, Torrance GW, et al. Multiattribute and singleattribute utility functions for the health utilities index mark 3 system. Med Care 2002;113 - 28. 19. Anonymous. S-Plus 2000. Data Analysis Products Division, MathSoft Inc:Seattle (WA); 1999.
208 Nichol et al
20. Anonymous. SOLAS for missing data analysis. 3.0 ed. Statistical Solutions Inc. Boston (MA); 2000. 21. Smith DM. OSWALD. 3.4 ed. Lancaster University Department of Mathematics and Statistics: Lancaster; 2000. 22. TreeAge Software I. Decision Analysis by TreeAge. DATA for Windows. Version 3.5 User’s Manual. Boston, MA; 1999. 23. Barber JA, Thompson SG. Analysis and interpretation of cost data in randomised controlled trials: review of published studies. BMJ 1998;1195 - 2000. 24. Briggs AH, O’Brien BJ. The death of cost-minimization analysis? Health Econ Lett 2000;3 - 10. 25. Etzioni RD, Feuer EJ, Sullivan SD, et al. On the use of survival analysis techniques to estimate medical care costs. J Health Econ 1999;365 - 80. 26. Hallstrom AP, Sullivan SD. On estimating costs for economic evaluation in failure time studies. Med Care 1998;433 - 6. 27. Little RJA, Rubin DB. Statistical analysis with missing data. In: Watson GS, editor. Wiley series in probability and mathematical statistics. John Wiley and Sons: New York; 1987. 28. Rubin DB. Multiple imputation for nonresponse in surveys. Wiley: New York; 1987. 29. Graham JW, Schafer JL. On the performance of multiple imputation for multivariate data with small sample size. In: Hoyle RH, editor. Statistical strategies for small sample research. SAGE Publications: London; 1999. p. 1 - 32. 30. Efron B, Tibshirani RJ. An introduction to the bootstrap. Chapman and Hall: New York; 1993. 31. Sonnenberg FA, Beck JR. Markov models in medical decision making: a practical guide. Med Decis Making1993;322 - 38. 32. Naimark D, Krahn MD, Naglie G, et al. Primer on medical decision analysis: part 5-working with Markov processes. Med Decis Making 1997;152 - 9. 33. O’Brien BJ, Connolly SJ, Goeree R, et al. Cost-effectiveness of the implantable cardioverter-defibrillator: results from the Canadian Implantable Defibrillator Study (CIDS). Circulation 2001;1416 - 21. 34. Briggs A, Sculpher M, Buxton M. Uncertainty in the economic evaluation of health care technologies: the role of sensitivity analysis. Health Econ 1994;95 - 104. 35. Doubilet P, Begg CB, Weinstein MC, et al. Probabilistic sensitivity analysis using Monte Carlo simulation. A practical approach. Med Decis Making 1985;157 - 77. 36. Critchfield GC, Willard KE. Probabilistic analysis of decision trees using Monte Carlo simulation. Med Decis Making 1986;86 - 92.
American Heart Journal August 2005
37. van Hout BA, al Maiwenn J, Gordon GS, et al. Costs, effects and C/E ratios alongside a clinical trial. Health Econ 1994;309 - 19. 38. Dowie J. Clinical trials and economic evaluations? No, there are only evaluations. Health Econ 1997;87 - 9. 39. Pell JP, Sirel JM, Marsden AK, et al. Potential impact of public access defibrillators on survival after out of hospital cardiopulmonary arrest: retrospective cohort study. BMJ 2002;515. 40. Anonymous. Guidelines 2000 for cardiopulmonary resuscitation and emergency cardiovascular care. International consensus on science. Circulation 2000;1 - 291. 41. Nichol G, Hallstrom A, Ornato JP, et al. Potential cost-effectiveness of public access defibrillation in the United States. Circulation 1998;1315 - 20. 42. Nichol G, Laupacis A, Stiell I, et al. A cost-effectiveness analysis of potential improvements to emergency medical services for victims of out-of-hospital cardiac arrest. Ann Emerg Med 1996;711 - 20. 43. Groeneveld PW, Kwong JL, Liu Y, et al. Cost-effectiveness of automated external defibrillators on airlines. JAMA 2001;1482 - 9. 44. Valenzuela TD, Roe DJ, Nichol G, et al. Outcomes of rapid defibrillation by security officers after cardiac arrest in casinos. N Engl J Med 2000;1206 - 9. 45. Caffrey SL, Willoughby PJ, Pepe PE, et al. Public use of automated external defibrillators. N Engl J Med 2002;1242 - 7. 46. Bloom BS, Fendrick AM. Timing and timeliness in medical care evaluation. PharmacoEconomics 1996;183 - 7. 47. Boyle M, Furlong W, Torrance G, et al. Reliability of the health utilities index—Mark III used in the 1991 cycle 6 Canadian General Social Survey Health Questionnaire. Qual Life Res 1995;249 - 57. 48. Nichol G, Stiell IG, Hebert P, et al. What is the quality of life of survivors of cardiac arrest? A prospective study. Acad Emerg Med 1999;95 - 102. 49. Rubin DB, Schenker N. Multiple imputation for interval estimation from simple random samples with ignorable nonresponse. J Am Stat Assoc 1986;366 - 74. 50. Hsiao WC, Braun P, Dunn D, et al. Resource-based relative values: an overview. JAMA 1988;2347 - 53. 51. Hsiao WC, Braun P, Dunn D, et al. Results and policy implications of the resource based relative-value study. N Engl J Med 1988;881 - 8. 52. Anonymous. 2001 drug topics red bookMedical Economics Data Inc: New York; 2001. p. 800.