Biomedical subjects
A R Feinstein
Publications and source records attributed to A R Feinstein.
Beyond statistics: what is really important in medicine?
Explore the source record for details and available documents.
Monte Carlo methods in clinical research: applications in multivariable analysis.
BACKGROUND: Monte Carlo methods use "simulated" analyses with random numbers for solving problems, particularly those that defy solutions using mathematical theory alone. Research using Monte Carlo simulations is very popular in many branches of science and is sometimes done in clinical investigation. The origins and basic strategy of the technique, however, may not be well known to clinical researchers. The purpose of this paper is to describe the history and general principles of Monte Carlo methods and to demonstrate how Monte Carlo simulations were recently applied to examine a phenomenon in multivariable statistical analysis called the number of outcome events per independent variable (EPV). For example, in a cohort of 200 people, with 50 deaths and 5 independent (predictor) variables, EPV = 50/5 = 10. METHODS: The "real-world" data came from a clinical trial of 673 patients in which 7 variables were cogent predictors of 252 deaths, so that EPV = 252/7 = 36. For the Monte Carlo simulations, special models were used while allowing simulations of proportional hazards and logistic regression to maintain the basic relationship of variables and the same size of the original population, at EPV values of 2, 5, 10, 15, 20, and 25. RESULTS: The Monte Carlo simulations confirmed a previously undocumented "rule of thumb" stating that when the EPV is less than 10-20, the algebraic models used in logistic regression and proportional hazards regression may be unreliable, leading to imprecise or spurious results. CONCLUSION: Monte Carlo techniques offer attractive methods for clinical investigators to use in solving problems that are not amenable to customary mathematical approaches.
Context bias. A problem in diagnostic radiology.
OBJECTIVE: To determine whether radiologists' interpretations of images are biased by their context and by prevalence of disease in other recently observed cases. METHODS: A test set of 24 right pulmonary arteriograms with a 33% prevalence of pulmonary emboli (PE) was assembled and embedded in 2 larger groups of films. Group A contained 16 additional arteriograms, all showing PE involving the right lung, so that total prevalence was 60%. Group B contained 16 additional arteriograms without PE so that total prevalence was 20%. Six radiologists were randomly assigned to see either group first and then "cross over" to review the other group after a hiatus of at least 8 weeks. The direction of changes in a 5-point rating scale for the 2 readings of each film in the test set was compared with the sign test; mean sensitivity, specificity, and areas under receiver operating characteristic (ROC) curves were compared with the paired t test. RESULTS: In the context of group A's higher disease prevalence, radiologists shifted more of their diagnoses toward higher suspicion than expected by chance (P=.03, sign test). In group A, mean sensitivity for diagnosing PE was significantly higher (75% vs 60%; P=.04), and area under the ROC curve was significantly larger (0.88 vs 0.82; P=.02). CONCLUSIONS: Radiologists' diagnoses are significantly influenced by the context of interpretation, even when spectrum and verification bias are avoided. This "context bias" effect is unique to the evaluation of subjectively interpreted tests, and illustrates the difficulty of obtaining unbiased estimates of diagnostic accuracy for both new and existing technologies.
Meta-analysis and meta-analytic monitoring of clinical trials.
Randomized trials are effective and usually unbiased for showing the average results in a selected outcome variable for treatment A versus treatment B, and meta-analyses produce an average of these averages. The results of both the trials and meta-analyses are often pragmatically unsatisfactory, however, because they do not reflect cogent distinctions desired by practising clinicians in the heterogeneous subgroups formed by diverse components in the patients' baseline states, in proficiency of therapy, and in additional outcome phenomena. If the inadequacies of previous trials have led to performance of a suitable new trial, it should not be stopped by the numbers emerging from meta-analyses of prior non-pertinent results.
Clinical symptoms and comorbidity: significance for the prognostic classification of cancer.
BACKGROUND: In 1992, the Cancer Registries Amendment Act allotted 30 million dollars annually for five years to establish a national program of cancer registries. Because the cornerstone of cancer staging is the Tumor, Node, Metastasis (TNM) system, this article is devoted to a brief history of the system, to important concepts of clinical biology that should be included in classification systems for cancer, and to sources and potential solutions for current problems. METHODS: A qualitative review of published literature on cancer staging, prognosis, and treatment effectiveness was performed, notes from previous American Joint Committee on Cancer (AJCC) meetings were reviewed, and a discussion with a former AJCC member was completed. RESULTS: Despite an excellent description of a tumor's size and extent of anatomic spread, the TNM system does not alone account for the cancer's clinical biology which is manifested by both the structural form of a tumor its physiological function in a patient. Important prognostic information can be determined by a patient's symptoms, which reflect some of a tumor's biologic behavior, and by comorbidity that is not a feature of the cancer itself. Five reasons were identified for the adherence to a strictly morphologic staging system. CONCLUSIONS: Widespread use of the TNM system during the past 30 years has unquestionably helped to standardize the classification of cancer and to improve prognostic estimates. Nevertheless, the estimates remain relatively imprecise, impairing the evaluation of treatment effectiveness. A prime scientific challenge in current cancer staging is to incorporate the omitted patient-based variables to produce an improved clinical system of classification.
Twentieth century paradigms that threaten both scientific and humane medicine in the twenty-first century.
Explore the source record for details and available documents.
A simulation study of the number of events per variable in logistic regression analysis.
We performed a Monte Carlo study to evaluate the effect of the number of events per variable (EPV) analyzed in logistic regression analysis. The simulations were based on data from a cardiac trial of 673 patients in which 252 deaths occurred and seven variables were cogent predictors of mortality; the number of events per predictive variable was (252/7 =) 36 for the full sample. For the simulations, at values of EPV = 2, 5, 10, 15, 20, and 25, we randomly generated 500 samples of the 673 patients, chosen with replacement, according to a logistic model derived from the full sample. Simulation results for the regression coefficients for each variable in each group of 500 samples were compared for bias, precision, and significance testing against the results of the model fitted to the original sample. For EPV values of 10 or greater, no major problems occurred. For EPV values less than 10, however, the regression coefficients were biased in both positive and negative directions; the large sample variance estimates from the logistic model both overestimated and underestimated the sample variance of the regression coefficients; the 90% confidence limits about the estimated values did not have proper coverage; the Wald statistic was conservative under the null hypothesis; and paradoxical associations (significance in the wrong direction) were increased. Although other factors (such as the total number of events, or sample size) may influence the validity of the logistic model, our findings indicate that low EPV can lead to major problems.
Two centuries of conflict-collaboration between medicine and mathematics.
Explore the source record for details and available documents.
Epidemiologic and clinical challenges in reviving the necropsy.
Explore the source record for details and available documents.
Re: "Epidemiology of gallstone disease in Italy: prevalence data of the Multicenter Italian Study on Cholelithiasis (M.I.COL.)".
Explore the source record for details and available documents.
Meta-analysis: statistical alchemy for the 21st century.
Explore the source record for details and available documents.
The etymology and pathogenesis of "wonk".
Explore the source record for details and available documents.
Importance of events per independent variable in proportional hazards regression analysis. II. Accuracy and precision of regression estimates.
The analytical effect of the number of events per variable (EPV) in a proportional hazards regression analysis was evaluated using Monte Carlo simulation techniques for data from a randomized trial containing 673 patients and 252 deaths, in which seven predictor variables had an original significance level of p < 0.10. The 252 deaths and 7 variables correspond to 36 events per variable analyzed in the full data set. Five hundred simulated analyses were conducted for these seven variables at EPVs of 2, 5, 10, 15, 20, and 25. For each simulation, a random exponential survival time was generated for each of the 673 patients, and the simulated results were compared with their original counterparts. As EPV decreased, the regression coefficients became more biased relative to the true value; the 90% confidence limits about the simulated values did not have a coverage of 90% for the original value; large sample properties did not hold for variance estimates from the proportional hazards model, and the Z statistics used to test the significance of the regression coefficients lost validity under the null hypothesis. Although a single boundary level for avoiding problems is not easy to choose, the value of EPV = 10 seems most prudent. Below this value for EPV, the results of proportional hazards regression analyses should be interpreted with caution because the statistical model may not be valid.
Importance of events per independent variable in proportional hazards analysis. I. Background, goals, and general strategy.
Multivariable methods of analysis can yield problematic results if methodological guidelines and mathematical assumptions are ignored. A problem arising from a too-small ratio of events per variable (EPV) can affect the accuracy and precision of regression coefficients and their tests of statistical significance. The problem occurs when a proportional hazards analysis contains too few "failure" events (e.g., deaths) in relation to the number of included independent variables. In the current research, the impact of EPV was assessed for results of proportional hazards analysis done with Monte Carlo simulations in an empirical data set of 673 subjects enrolled in a multicenter trial of coronary artery bypass surgery. The research is presented in two parts: Part I describes the data set and strategy used for the analyses, including the Monte Carlo simulation studies done to determine and compare the impact of various values of EPV in proportional hazards analytical results. Part II compares the output of regression models obtained from the simulations, and discusses the implication of the findings.
Methodologic sources of inconsistent prognoses for post-acute myocardial infarction.
PURPOSE: To investigate basic methodologic problems that could explain inconsistent and contradictory results for predictor variables in studies of prognosis after myocardial infarction (MI). MATERIALS AND METHODS: Studies on postinfarct prognosis published in English between 1979 and 1991 were identified with a MEDLINE literature search. The key words used for the computer search were: "prognosis" and "myocardial infarction" in the title and "mortality" or "survival" or "outcome" in the title or abstract. Reference lists in the reports captured by the search were examined for pertinent articles, and additional articles were sought in the index pages of two prominent journals. To be included in the analysis, a study had to fulfill the following eligibility criteria: a cohort study or randomized, controlled trial; sample size > or = 50 patients; a clear identification of the time when follow-up began, after the acute phase of MI and either before or at hospital discharge; follow-up for a minimum of 6 months or median/mean of 1 year; and multivariable analysis for intervals no longer than 2 years after the MI. Eight methodologic standards addressing sources of major problems were established and applied to each study. RESULTS: Of 766 reports identified, 111 fulfilled the eligibility criteria. The median number of standards fulfilled was 3, the highest 6. The proportions of studies complying with each of the 8 methodologic standards were: (1) inception cohort, 60%; (2) total death as an unequivocal outcome, 54%; (3) verification of cause-specific deaths (in 62 studies analyzing cardiac death), 37%; (4) analysis of crucial variables describing baseline severity, 13%; (5) indication of quantitative scope of the spectrum of baseline severity, 20%; (6) reproducible classification of candidate predictor variables, 40%; (7) adequate identification of quantitative importance of and boundaries for statistically significant predictor variables, 39%; and (8) evaluation of impact of treatment on predictor variables, 13%. CONCLUSIONS: The results show that studies on postinfarct prognosis have frequently disregarded basic methodologic principles. Suitable adherence to these principles in future research will allow improved interpretation of results and can reduce inconsistent findings, while improving the applicability of the identified predictors.
Complications of peripheral arteriography: a new system to identify patients at increased risk.
PURPOSE: The most quoted literature on arteriographic complications is based on self-reports collected during the mid 1970s. We sought to determine whether those results remain valid despite changes in arteriographic practice and whether patient subgroups at increased risk could be identified. METHODS: Five hundred forty-nine consecutive patients were examined after arteriography and twice over 72 hours. Patients were telephoned at least 2 weeks later to identify delayed complications. The sample was divided into two groups to allow independent validation of suspected prognostic factors. RESULTS: The rate of major complications was 2.9% (16/549), but varied from 0.7% to 9.1% among three strata of relative risk. Rates were highest in patients studied for suspected aortic dissection, mesenteric ischemia, gastrointestinal bleeding, or symptomatic carotid artery stenosis and lowest in patients with trauma or aneurysmal disease. Patients studied for claudication or limb-threatening ischemia had intermediate risk (2.0%). Within these strata, congestive heart failure and furosemide use were the only variables independently associated with a significantly increased complication rate. CONCLUSIONS: Previous reports have overestimated the risk of arteriography for trauma or aneurysm but substantially underestimate the risk for patients with other common conditions. Such stratified complication rates are essential to understand relative costs and benefits of arteriography and other vascular imaging modalities in specific clinical situations.
The crisis in clinical research. Essay review.
Explore the source record for details and available documents.