Effect of nesiritide in patients with acute decompensated heart failure.

BACKGROUND: Nesiritide is approved in the United States for early relief of dyspnea in patients with acute heart failure. Previous meta-analyses have raised questions regarding renal toxicity and the mortality associated with this agent. METHODS: We randomly assigned 7141 patients who were hospitalized with acute heart failure to receive either nesiritide or placebo for 24 to 168 hours in addition to standard care. Coprimary end points were the change in dyspnea at 6 and 24 hours, as measured on a 7-point Likert scale, and the composite end point of rehospitalization for heart failure or death within 30 days. RESULTS: Patients randomly assigned to nesiritide, as compared with those assigned to placebo, more frequently reported markedly or moderately improved dyspnea at 6 hours (44.5% vs. 42.1%, P=0.03) and 24 hours (68.2% vs. 66.1%, P=0.007), but the prespecified level for significance (P≤0.005 for both assessments or P≤0.0025 for either) was not met. The rate of rehospitalization for heart failure or death from any cause within 30 days was 9.4% in the nesiritide group versus 10.1% in the placebo group (absolute difference, -0.7 percentage points; 95% confidence interval [CI], -2.1 to 0.7; P=0.31). There were no significant differences in rates of death from any cause at 30 days (3.6% with nesiritide vs. 4.0% with placebo; absolute difference, -0.4 percentage points; 95% CI, -1.3 to 0.5) or rates of worsening renal function, defined by more than a 25% decrease in the estimated glomerular filtration rate (31.4% vs. 29.5%; odds ratio, 1.09; 95% CI, 0.98 to 1.21; P=0.11). CONCLUSIONS: Nesiritide was not associated with an increase or a decrease in the rate of death and rehospitalization and had a small, nonsignificant effect on dyspnea when used in combination with other therapies. It was not associated with a worsening of renal function, but it was associated with an increase in rates of hypotension. On the basis of these results, nesiritide cannot be recommended for routine use in the broad population of patients with acute heart failure. (Funded by Scios; ClinicalTrials.gov number, NCT00475852.).

Epoetin treatment of anemia associated with cancer therapy: a systematic review and meta-analysis of controlled clinical trials.

Epoetin treatment offers an attractive but costly alternative to red blood cell transfusion for managing anemia associated with cancer therapy. The goal of this review is to facilitate more efficient use of epoetin by 1) quantifying the effects of epoetin on the likelihood of transfusion and on quality of life in patients with cancer treatment-related anemia and 2) evaluating whether outcomes are superior when epoetin treatment is initiated at higher hemoglobin thresholds. Two independent reviewers followed a prospective protocol for identifying studies. Outcomes data were combined with the use of a random-effects meta-analysis model. Double-blind, randomized, controlled trials that minimized patient exclusions were defined as higher quality for sensitivity analysis; randomized but unblinded trials and trials with excessive exclusions were included in the meta-analysis but were defined as lower quality. Twenty-two trials (n = 1927) met inclusion criteria, and 12 (n = 1390) could be combined for estimation of odds of transfusion. Epoetin decreased the percentage of patients transfused by 9%-45% in adults with mean baseline hemoglobin concentrations of 10 g/dL or less (seven trials; n = 1080), by 7%-47% in those with hemoglobin concentrations greater than 10 g/dL but less than 12 g/dL (seven trials; n = 431), and by 7%-39% in those with hemoglobin concentrations of 12 g/dL or higher (five trials; n = 308). In sensitivity analysis, the combined odds ratio for transfusion in epoetin-treated patients as compared with controls was 0.45 (95% confidence interval [CI] = 0.33 to 0.62) in higher quality studies and 0.14 (95% CI = 0.06 to 0.31) in lower quality studies. The number of patients needed to treat to prevent one transfusion is 4.4 for all studies, 5.2 for higher quality studies, and 2.6 for lower quality studies. Only studies with mean baseline hemoglobin concentrations of 10 g/dL or less reported statistically significant effects of epoetin treatment on quality of life; quality-of-life data were insufficient for meta-analysis. No studies addressed epoetin's effects on anemia-related symptoms. We conclude that epoetin reduces the odds of transfusion for cancer patients undergoing therapy. Evidence is insufficient to determine whether initiating epoetin earlier spares more patients from transfusion or results in better quality of life than waiting until hemoglobin concentrations decline to nearly 10 g/dL.

Clinical outcomes of bivalirudin for ischemic heart disease.

BACKGROUND: Current treatment strategies for percutaneous coronary revascularization and acute coronary syndromes incorporate thrombin inhibition with either unfractionated or fractionated heparin. The peptide bivalirudin (Hirulog) is a direct thrombin inhibitor whose pharmacological properties differ from those of heparin. We conducted a systematic overview (meta-analysis) to assess the effect of bivalirudin on 4 end points: death, myocardial infarction, major hemorrhage, and the composite of death or infarction. METHODS AND RESULTS: Six trials (5674 patients) represent the randomized, controlled bivalirudin experience, including 4603 patients undergoing elective percutaneous coronary revascularization and 1071 patients with acute coronary syndromes. ORs for the 4 clinical end points were calculated for each trial. Four trials (4973 patients) that compared bivalirudin with heparin were combined with the use of a random-effects model. In these trials, bivalirudin was associated with a significant reduction in the composite of death or infarction (OR 0.73, 95% CI 0.57 to 0.95; P=0.02) at 30 to 50 days, or 14 fewer events per 1000 patients so treated. There also was a significant reduction in major hemorrhage for the same trials (OR 0.41, 95% CI 0. 32 to 0.52; P<0.001, or 58 fewer events per 1000 patients so treated). A similar analysis combined 2 dose-ranging trials (701 patients) that compared therapeutic (activated partial thromboplastin time more than twice the control time) with subtherapeutic bivalirudin anticoagulation (activated partial thromboplastin time less than twice the control time). CONCLUSIONS: Bivalirudin is at least as effective as heparin, with clearly superior safety. Thus, it provides an unprecedented net clinical benefit over heparin in patients with ischemic heart disease.

Benefit of glycoprotein IIb/IIIa inhibition in patients with acute coronary syndromes and troponin t-positive status: the paragon-B troponin T substudy.

BACKGROUND: Troponin T (TnT) is valuable for short- and long-term risk stratification of patients with acute coronary syndromes (ACS). It also may predict which ACS patients will benefit from glycoprotein (GP) IIb/IIIa blockade. METHODS AND RESULTS: We prospectively studied 1160 patients with non-ST-segment elevation ACS randomized in PARAGON-B to receive lamifiban, an intravenous GP IIb/IIIa antagonist, or placebo. TnT levels were obtained before study treatment began and 24 to 72 hours later; assays were performed by a blinded core laboratory. At baseline, 40.2% of patients were TnT-positive (>/=0.1 ng/mL); these patients were older and more often male or smokers. Patients positive at baseline had a significantly higher rate of the primary end point (composite of death, myocardial [re]infarction, or severe recurrent ischemia at 30 days; odds ratio, 1.5; 95% CI, 1.1 to 2.1) than those who were TnT-negative. Lamifiban was associated with significant reduction in the primary end point (from 19.4% to 11.0%, P=0.01) among TnT-positive patients but not among TnT-negative patients (11.2% for placebo versus 10.8% for lamifiban, P=0.86; P=0.08 for test of interaction between TnT status and treatment assignment). This pattern held for the end points of death alone and death or myocardial (re)infarction at 30 days. Peak TnT level at 48 hours did not differ with lamifiban treatment. CONCLUSIONS: TnT predicts poor short-term outcomes in non-ST-segment elevation ACS. Treatment benefit with lamifiban is limited almost exclusively to TnT-positive patients, reducing 30-day adverse outcomes to a rate nearly identical to that of negative patients.

Reactivation of ischemic events in acute coronary syndromes: results from GUSTO-IIb. Gobal Use of Strategies To Open occluded arteries in acute coronary syndromes.

OBJECTIVES: We sought to determine the incidence of and risk factors for thrombotic events early after discontinuing antithrombin therapy in patients with acute coronary syndromes. BACKGROUND: Discontinuation of treatment with heparin and other thrombin inhibitors in patients with unstable coronary syndromes has related to clinical and biochemical evidence of early reactivation of thrombosis. METHODS: We studied 8,943 of the 12,142 patients with acute coronary syndromes enrolled in the Global Use of Strategies To Open occluded arteries in acute coronary syndromes trial of hirudin versus heparin. We excluded patients who received no study drug, lacked timing data, died or had myocardial (re)infarction [(re)MI] during study-drug infusion, or began heparin treatment within 2 h after treatment with the study drug was stopped. We assessed the incidence and timing of (re)MI by type and timing of antithrombin treatment. RESULTS: In all, 215 patients (2.4%) suffered (re)MI, 49 within 12 h of antithrombin therapy discontinuation and 166 between hour 12 and hospital discharge. The duration of infusion did not differ between the hirudin and heparin groups. The rate of early re(MI) after drug therapy discontinuation was significantly higher in patients given heparin versus hirudin (0.8% vs. 0.3%, p = 0.002). Patients with (re)MI had higher mortality at 30 days (23.6% vs. 2.4%, p = 0.001) and 1 year (35.2% vs. 6.7%, p = 0.001) compared with patients without (re)MI. CONCLUSIONS: The incidence of (re)MI was clustered within 12 h of heparin therapy discontinuation, with the greatest risk within 4 h. There was no evidence of early reactivation of thrombotic events after hirudin. Patients who had (re)infarction had worse outcomes. Better understanding of the mechanism and possible prevention of recurrent thrombosis is needed.

A randomized, placebo-controlled trial of enoxaparin after high-risk coronary stenting: the ATLAST trial.

OBJECTIVES: We performed a multicenter, double-blind placebo-controlled trial to examine the efficacy and safety of enoxaparin in patients at high risk for stent thrombosis (ST). BACKGROUND: The optimal antithrombotic regimen for such patients is unknown. METHODS: We randomized 1,102 patients with clinical, angiographic or ultrasonographic features associated with an increased risk of ST to receive either twice-daily injections of weight-adjusted enoxaparin or placebo for 14 days after stenting. All patients received aspirin and ticlopidine. The primary end point was a 30-day composite end point of death, myocardial infarction (MI) or urgent revascularization. RESULTS: The target enrollment for the study was 2,000 patients. However, the trial was terminated prematurely at 1,102 patients after interim analysis revealed an unexpectedly low event rate. The primary outcome occurred in 1.8% enoxaparin-treated patients versus 2.7% treated with placebo (odds ratio [OR] 0.66; 95% confidence interval [CI] 0.29 to 1.5, p = 0.30); for death or MI the rates were 0.9% vs. 2.2%, respectively (OR 0.41, 95% CI 0.14 to 1.2, p =0.13); and for MI, 0.4% vs. 1.6%, respectively (OR 0.22, 95% CI 0.05 to 0.99, p = 0.04). The groups had comparable rates of major bleeding (3.3% for enoxaparin, 1.6% for placebo, p =0.08), but minor nuisance bleeding was increased with enoxaparin (25% vs. 5.1%, p < 0.001). CONCLUSIONS: The clinical outcomes of patients at increased risk of ST are more favorable than previously reported, rendering routine oral antiplatelet therapy adequate for most. However, given its relative safety and potential to reduce the risk of subsequent infarction, a 14-day course of enoxaparin may be considered for carefully selected patients.

Reduction in heart failure events by the addition of a clinical pharmacist to the heart failure management team: results of the Pharmacist in Heart Failure Assessment Recommendation and Monitoring (PHARM) Study.

BACKGROUND: The multidisciplinary approach to managing heart failure has been shown to improve outcomes. The role of a clinical pharmacist in treating heart failure has not been evaluated. METHODS: One hundred eighty-one patients with heart failure and left ventricular dysfunction (ejection fraction <45) undergoing evaluation in clinic were randomized to an intervention or a control group. Patients in the intervention group received clinical pharmacist evaluation, which included medication evaluation, therapeutic recommendations to the attending physician, patient education, and follow-up telemonitoring. The control group received usual care. The primary end point was combined all-cause mortality and heart failure clinical events. All clinical events were adjudicated by a blinded end point committee. RESULTS: Baseline characteristics were similar except for slightly higher age in the intervention group. Median follow-up was 6 months. All-cause mortality and heart failure events were significantly lower in the intervention group compared with the control group (4 vs 16; P= .005). In addition, patients in the intervention group received higher angiotensin-converting enzyme inhibitor doses as reflected by the median fraction of target reached (25th and 75th percentiles), 1.0 (0.5 and 1) and 0.5 (0.1875 and 1) in the intervention and control groups, respectively (P<.001). The use of other vasodilators in angiotensin-converting enzyme inhibitor-intolerant patients was higher in the intervention group (75% vs 26%; P= .02). CONCLUSIONS: Outcomes in heart failure can be improved with a clinical pharmacist as a member of the multidisciplinary heart failure team. This observation may be due to higher doses of angiotensin-converting enzyme inhibitors and/or closer follow-up.

Comparison and meta-analysis of randomized trials of endarterectomy for symptomatic carotid artery stenosis.

OBJECTIVE: Comparison and meta-analysis of randomized trials of carotid endarterectomy for symptomatic stenosis of the extracranial carotid artery. BACKGROUND: Randomized trials (North American Symptomatic Carotid Endarterectomy Trial [NASCET], the European Carotid Surgery Trial [ECST], and the VA Cooperative Study [VACS]) each show that carotid endarterectomy improves outcomes in selected symptomatic patients with high-grade extracranial carotid artery stenosis. Direct comparisons among the studies have not been possible because of differing methodologies, endpoints, and formats of data reporting. DESIGN/METHODS: Data for specified endpoints and corresponding person-years at risk were obtained for each trial. The rates of nonfatal stroke, nonfatal myocardial infarction, or death for surgically or medically treated patients in the perioperative period (30 days) and thereafter were compared (both including and excluding perioperative events) and then combined using meta-analytic techniques. Data from NASCET and ECST were also analyzed for differences in outcomes by sex. RESULTS: Event rate estimates (with 95% confidence intervals [95% CI]) for the first 30 days (events per person-year, primarily nonfatal stroke) for medically treated patients were 0.44 (0.22 to 0.76) for NASCET, 0.15 (0.04 to 0.38) for ECST, and 0.27 (0.03 to 0.96) for VACS. For surgically treated patients, the corresponding rates (per person-year) were 0.78 (0.49 to 1.19), 0.63 (0.41 to 0.94), and 1.27 (0.58 to 2.43). Event rates per year after 30 days were higher for medically treated patients (0.20 [0.16 to 0.25] versus 0.08 [0.05 to 0.11] for NASCET; 0.12 [0.10 to 0.15] versus 0.07 [0.06 to 0.09] for ECST; and 0.15 [0.07 to 0.25] versus 0.07 [0.03 to 0.16] for VACS). There were no significant differences among the trials, with an overall benefit for surgical therapy (risk ratio estimate, RR = 0.67, 95% CI = 0.54 to 0.83). There were no significant sex-based differences between NASCET and ECST and the overall benefit was not significantly different for men and women (RR = 0.58, 95% CI = 0.45 to 0.74 for men; RR = 0.84, 95% CI = 0.57 to 1.25 for women). CONCLUSIONS: Adjusting for primary endpoints and duration of follow-up, carotid endarterectomy has a similar benefit for symptomatic patients across trials and a similar benefit for men and women.

Initial experience with factor-Xa inhibition in percutaneous coronary intervention: the XaNADU-PCI Pilot.

BACKGROUND: Direct factor (F)Xa inhibition is an attractive method to limit thrombotic complications during percutaneous coronary intervention (PCI). OBJECTIVES: To investigate drug levels achieved, effect on coagulation markers, and preliminary efficacy and safety of several doses of DX-9065a, an intravenous, small molecule, direct, reversible FXa inhibitor during PCI. PATIENTS AND METHODS: Patients undergoing elective, native-vessel PCI (n = 175) were randomized 4 : 1 to open-label DX-9065a or heparin in one of four sequential stages. DX-9065a regimens in stages I-III were designed to achieve concentrations of > 100 ng mL-1, > 75 ng mL-1, and > 150 ng mL-1. Stage IV used the stage III regimen but included patients recently given heparin. RESULTS: At 15 min median (minimum) DX-9065a plasma levels were 192 (176), 122 (117), 334 (221), and 429 (231) ng mL-1 in stages I-IV, respectively. Median whole-blood international normalized ratios (INRs) were 2.6 (interquartile range 2.5, 2.7), 1.9 (1.8, 2.0), 3.2 (3.0, 4.1), and 3.8 (3.4, 4.6), and anti-FXa levels were 0.36 (0.32, 0.38), 0.33 (0.26, 0.39), 0.45 (0.41, 0.51), and 0.62 (0.52, 0.65) U mL-1, respectively. Stage II enrollment was stopped (n = 7) after one serious thrombotic event. Ischemic and bleeding events were rare and, in this small population, showed no clear relation to DX-9065a dose. CONCLUSIONS: Elective PCI is feasible using a direct FXa inhibitor for anticoagulation. Predictable plasma drug levels can be rapidly obtained with double-bolus and infusion DX-9065a dosing. Monitoring of DX-9065a may be possible using whole-blood INR. Direct FXa inhibition is a novel and potentially promising approach to anticoagulation during PCI that deserves further study.

Meta-analysis of screening and diagnostic tests.

Screening and diagnostic tests are common in the fields of psychology, medicine, and education. Often there are several competing tests, and decisions must be made about the relative accuracy of those tests. This article describes a general measure that can be used for both continuous and dichotomous outcome measures. It is the standardized distance between the means of the 2 populations. For continuous measures, it is the effect size measure. For dichotomous measures, it is proportional to the logarithm of the odds of the sensitivity plus the logarithm of the odds of the specificity. The measure is easily computed for both kinds of outcomes. Properties of this measure are discussed, and examples are given. Ths use of this measure to compare the average performance of different tests is described.

A metaanalysis of predischarge risk stratification after acute myocardial infarction with stress electrocardiographic, myocardial perfusion, and ventricular function imaging.

We assessed the relation of abnormal predischarge non-invasive test results to outcomes in postmyocardial infarction patients. We included series published from 1980 to 1995 containing only myocardial infarction patients, enrolling most patients after 1980, testing within 6 weeks of infarction, having follow-up rates > 80%, and having 2 x 2 frequency outcome rates for test results, that were the latest of multiple reports. Sensitivity, specificity, and predictive values were calculated for test results for 1-year outcomes (cardiac death, cardiac death or reinfarction). Univariable and summary odds were calculated for test results. Reports (n = 54) included a total of 19,874 patients and were primarily retrospective (76%) and small series (35% of reports included < 5 deaths). One-year mortality ranged from 2.5% for pharmacologic stress echocardiography to 9.3% for exercise radionuclide angiography. Positive predictive values for most noninvasive risk markers were < 0.10 for cardiac death and < 0.20 for death or reinfarction. Electrocardiographic, symptomatic, and scintigraphic risk markers of ischemia (ST-segment depression, angina, a reversible defect) were less sensitive (< or = 44%) for identifying morbid and fatal outcomes than markers of left ventricular dysfunction or heart failure (exercise duration, impaired systolic blood pressure response, and peak left ventricular ejection fraction). The positive predictive value of predischarge noninvasive testing is low. Markers of left ventricular dysfunction appear to be better predictors than markers of ischemia. Limitations of the literature-small samples and widely varying event rates-impede our ability to discern the accuracy of pre-discharge noninvasive testing. More rigorous, controlled trials are required to elucidate the relative value of these tests for risk stratification.

Prognostic usefulness of the six-minute walk in patients with advanced congestive heart failure secondary to ischemic or nonischemic cardiomyopathy.

Clinicians have relied on history and results from physical examinations to guide treatment of patients with advanced congestive heart failure, but these results may not reflect disease severity or hemodynamic status. We assessed how the distance walked in 6 minutes relates to clinical outcomes and symptoms of such patients. We compared the rates of death, hospitalization, and their composite at 1 year by the distance walked in 6 minutes at baseline and at 1 month, and by the change in distance between baseline and 1 month in 440 patients enrolled in a randomized trial. We also assessed the relations of baseline distance walked to symptom score and New York Heart Association class. The median distance increased from 218 m at baseline to 280 m at 1 month. Of 365 patients able to perform the baseline walk, 121 (33%) died and 217 (60%) were hospitalized compared with 46 (61%) and 34 (45%) of 75 patients unable to walk at baseline. Baseline distance significantly predicted mortality (hazard ratio 0.58/100-m increase, 95% confidence interval 0.50 to 0.68, p <0.001), even after adjustment. Baseline distance also significantly predicted hospitalization and the composite end point, as did the 1-month distance walked. The change in distance walked from baseline to 1 month did not predict any end point. Baseline distance correlated only moderately with symptom score (r = -0.385, p <0.001) and New York Heart Association class (r = -0.468, p <0.001). Distance walked during 6 minutes independently and strongly predicts mortality and hospitalization in patients with advanced congestive heart failure. This may be a simple, noninvasive, objective way to risk-stratify these patients and standardize their treatment.

Hormone replacement therapy and the risk of colorectal cancer: a meta-analysis.

OBJECTIVE: To review systematically the association between hormone replacement therapy (HRT) and the risk of developing or dying from colorectal cancer. DATA SOURCES: We searched the English-language literature using MEDLINE, Current Contents, CancerLit, and bibliographies of selected studies. METHODS OF STUDY SELECTION: We included studies that specifically addressed the association of HRT with colorectal cancer, had adequate controls, and had retrievable risk estimates. We excluded letters, reviews, and multiple publications of the same data. TABULATION, INTEGRATION, AND RESULTS: Studies were evaluated independently by two of the authors. The exposures of interest were ever, recent, or former use of HRT, and the main outcome measures were colon and rectal cancer incidence and mortality. To reduce the risk of a "healthy estrogen user" bias, we defined recent HRT use as either at time of assessment or within the previous year. The most adjusted risk estimates were extracted. We used a random-effects model to calculate summary relative risks (RRs) and confidence intervals (CIs). Recent use of HRT was associated with a 33% reduction in the risk of colon cancer (RR = 0.67; 95% CI 0.59, 0.77). Protection was limited to recent users; the risk of colon cancer with ever use of HRT was 0.92 (95% CI 0.79, 1.08). Duration of use was not significant. Three studies addressed the risk of fatal colon cancer; the summary RR for death from colon cancer in HRT users was 0.72 (95% CI 0.64, 0.81) compared with nonusers. Rectal cancer incidence was not associated with HRT. CONCLUSION: The risk of colon cancer may be decreased among recent postmenopausal HRT users. Although data are limited, the risk of fatal colon cancer also may be lower in HRT users.

The Agency for Health Care Policy and Research Smoking Cessation Clinical Practice Guideline. Findings and implications for psychologists.

Smoking is the leading cause of preventable morbidity and mortality in the United States, and the health benefits of quitting smoking are substantial. Nevertheless, over 25% of American adults (48 million individuals) continue to smoke, and the vast majority of quit attempts are unsuccessful. The Agency for Health Care Policy and Research recently addressed the smoking problem by conducting a 2-year research project that was published as the Smoking Cessation Clinical Practice Guideline (Fiore et al., 1996). This article reviews methods, analyses, and results from the Guideline project, and highlights major Guideline recommendations. Guideline findings and recommendations are discussed with respect to their implications for psychology.

Meta-analysis of multitreatment studies.

Studies comparing more than two competing therapies are common in several fields, but standard meta-analytic methods can make only pairwise comparisons. The methods proposed in this article, a generalization of current meta-analytic methods, allow for any number of competing therapies and include both fixed- and random-effects models.

Meta-analytic tools for medical decision making: a practical guide.

This paper is an extension of notes used for the short course in meta-analysis given at the 13th and 14th annual meetings of the Society for Medical Decision Making. The material covers both standard and evolving methods of meta-analysis. The methods include those for combining p-values, for analyzing general fixed-effects models, for analyzing contingency tables, and for analyzing count and continuous outcomes. For each general method, the authors present simplified formulas first, followed by more precise formulas when necessary. Similarly, both classic and Bayesian methods are presented where appropriate. Actual examples are used for methods.

An introduction to a Bayesian method for meta-analysis: The confidence profile method.

The Confidence Profile Method is a new Bayesian method that can be used to assess technologies where the available evidence involves a variety of experimental designs, types of outcomes, and effect measures; a variety of biases; combinations of biases and nested bases; uncertainty about biases; an underlying variability in the parameter of interest; indirect evidence; and technology families. The result of an analysis with the Confidence Profile Method is a posterior distribution for the parameter of interest, posterior distributions for other parameters, and a covariance matrix for all the parameters in the model. The posterior distributions incorporate all the uncertainty the assessor chooses to describe about any of the parameters used in the analysis.

Assessing uncertainty in cost-effectiveness analyses: application to a complex decision model.

A framework for quantifying uncertainty about costs, effectiveness measures, and marginal cost-effectiveness ratios in complex decision models is presented. This type of application requires special techniques because of the multiple sources of information and the model-based combination of data. The authors discuss two alternative approaches, one based on Bayesian inference and the other on resampling. While computationally intensive, these are flexible in handling complex distributional assumptions and a variety of outcome measures of interest. These concepts are illustrated using a simplified model. Then the extension to a complex decision model using the stroke-prevention policy model is described.

Predicting the cost of illness: a comparison of alternative models applied to stroke.

Predictions of cost over well-defined time horizons are frequently required in the analysis of clinical trials and social experiments, for decision models investigating the cost-effectiveness of interventions, and for macro-level estimates of the resource impact of disease. With rare exceptions, cost predictions used in such applications continue to take the form of deterministic point estimates. However, the growing availability of large administrative and clinical data sets offers new opportunities for a more general approach to disease cost forecasting: the estimation of multivariable cost functions that yield predictions at the individual level, conditional on intervention(s), patient characteristics, and other factors. This raises the fundamental question of how to choose the "best" cost model for a given application. The central purpose of this paper is to demonstrate how to evaluate competing models on the basis of predictive validity. This concept is operationalized according to three alternative criteria: 1) root mean square error (RMSE), for evaluating predicted mean cost; 2) mean absolute error (MAE), for evaluating predicted median cost; and 3) a logarithmic scoring rule (log score), an information-theoretic index for evaluating the entire predictive distribution of cost. To illustrate these concepts, the authors conducted a split-sample analysis of data from a national sample of Medicare-covered patients hospitalized for ischemic stroke in 1991 and followed to the end of 1993. Using test and training samples of about 500,000 observations each, they investigated five models: single-equation linear models, with and without log transform of cost; two-part (mixture) models, with and without log transform, to directly address the problem of zero-cost observations; and a Cox proportional-hazards model stratified by time interval. For deriving the predictive distribution of cost, the log transformed two-part and proportional-hazards models are superior. For deriving the predicted mean or median cost, these two models and the commonly used log-transformed linear model all perform about the same. The untransformed models are dominated in every instance. The approaches to model selection illustrated here can be applied across a wide range of settings.