The "resident's dilemma"? Values and strategies of medical residents for education interactions: a cellular automata simulation.

OBJECTIVES: Medical residents engage in formal and informal education interactions with fellow residents during the working day, and can choose whether to spend time and effort on such interactions. Time and effort spent on such interactions can bring learning and personal satisfaction to residents, but may also delay completion of clinical work. METHODS: Using hypothetical cases, we assessed the values and strategies of internal medicine residents at one hospital for both cooperative and non-cooperative education interactions with fellow residents. We then used these data and cellular automata models of two-person games to simulate repeated interactions between residents, and to determine which strategies resulted in greatest accrued value. We conducted sensitivity analyses on several model parameters, to test the robustness of dominant strategies to model assumptions. RESULTS: Twenty-nine of the 57 residents (50.9%) valued cooperation more than non-cooperation no matter what the other resident did during the current interaction. Similarly, thirty-six residents (63.2%) endorsed an unconditional always-cooperate strategy no matter what the other resident had done during their previous interaction. In simulations, an always-cooperate strategy accrued more value (776.42 value units) than an aggregate of strategies containing non-cooperation components (675.0 value units, p = 0.052). Only when the probability of strategy errors reached 50%, or when values were re-ordered to match those of a Prisoner's Dilemma, did non-cooperation-based strategies accrue the most value. CONCLUSIONS: Cooperation-based values and strategies were most frequent among our residents, and dominated in simulations of repeated education interactions between them.

Selection of predictor variables for pneumonia using neural networks and genetic algorithms.

BACKGROUND: Artificial neural networks (ANN) can be used to select sets of predictor variable that incorporate nonlinear interactions between variables. We used a genetic algorithm, with selection based on maximizing network accuracy and minimizing network input-layer cardinality, to evolve parsimonious sets of variables for predicting community-acquired pneumonia among patients with respiratory complaints. METHODS: ANN were trained on data from 1044 patients in a training cohort, and were applied to 116 patients in a testing cohort. Chromosomes with binary genes representing input-layer variables were operated on by crossover recombination, mutation, and probabilistic selection based on a fitness function incorporating both network accuracy and input-layer cardinality. RESULTS: The genetic algorithm evolved best 10-variable sets that discriminated pneumonia in the training cohort (ROC areas, 0.838 for selection based on average cross entropy (ENT); 0.954 for selection based on ROC area (ROC)), and in the testing cohort (ROC areas, 0.847 for ENT selection; 0.963 for ROC selection), with no significant differences between cohorts. Best variable sets based on the genetic algorithm using ROC selection discriminated pneumonia more accurately than variable sets based on stepwise neural networks (ROC areas, 0.954 versus 0.879, p = 0.030), or stepwise logistic regression (ROC areas, 0.954 versus 0.830, p = 0.000). Variable sets of lower cardinalities were also evolved, which also accurately discriminated pneumonia. CONCLUSION: Variable sets derived using a genetic algorithm for neural networks accurately discriminated pneumonia from other respiratory conditions, and did so with greater accuracy than variables derived using stepwise neural networks or logistic regression in some cases.

Entering the black box of neural networks.

OBJECTIVES: Artificial neural networks have proved to be accurate predictive instruments in several medical domains, but have been criticized for failing to specify the information upon which their predictions are based. We used methods of relevance analysis and sensitivity analysis to determine the most important predictor variables for a validated neural network for community-acquired pneumonia. METHODS: We studied a feed-forward, back-propagation neural network trained to predict pneumonia among patients presenting to an emergency department with fever or respiratory complaints. We used the methods of full retraining, weight elimination, constant substitution, linear substitution, and data permutation to identify a consensus set of important demographic, symptom, sign, and comorbidity predictors that influenced network output for pneumonia. We compared predictors identified by these methods to those identified by a weight propagation analysis based on the matrices of the network, and by logistic regression. RESULTS: Predictors identified by these methods were clinically plausible, and were concordant with those identified by weight analysis, and by logistic regression using the same data. The methods were highly correlated in network error, and led to variable sets with errors below bootstrap 95% confidence intervals for networks with similar numbers of inputs. Scores for variable relevance tended to be higher with methods that precluded network retraining (weight elimination) or that permuted variable values (data permutation), compared with methods that permitted retraining (full retraining) or that approximated its effects (constant and linear substitution). CONCLUSION: Methods of relevance analysis and sensitivity analysis are useful for identifying important predictor variables used by artificial neural networks.

Parametric three-way receiver operating characteristic surface analysis using mathematica.

Three-way receiver operating characteristic (ROC) surface analysis involves the calculation of a volume under an ROC surface (VUS), which is a measure of discriminatory accuracy of 2 diagnostic tests for 3 diseases. Nonparametric methods for calculating VUS and its standard error have been developed. The author presents the code for roc3D, a Mathematica computer program for performing parametric ROC surface analysis. roc3D calculates VUS assuming a multinormal distribution of test results in the 3 diseased populations, provides user-specified pointwise confidence limits for VUS, and displays a 3-dimensional plot of the ROC surface. Limitations of the roc3D program are discussed.

Enhancement of clinicians' diagnostic reasoning by computer-based consultation: a multisite study of 2 systems.

CONTEXT: Computer-based diagnostic decision support systems (DSSs) were developed to improve health care quality by providing accurate, useful, and timely diagnostic information to clinicians. However, most studies have emphasized the accuracy of the computer system alone, without placing clinicians in the role of direct users. OBJECTIVE: To explore the extent to which consultations with DSSs improve clinicians' diagnostic hypotheses in a set of diagnostically challenging cases. DESIGN: Partially randomized controlled trial conducted in a laboratory setting, using a prospective balanced experimental design in 1995-1998. SETTING: Three academic medical centers, none of which were involved in the development of the DSSs. PARTICIPANTS: A total of 216 physicians: 72 at each site, including 24 internal medicine faculty members, 24 senior residents, and 24 fourth-year medical students. One physician's data were lost to analysis. INTERVENTION: Two DSSs, ILIAD (version 4.2) and Quick Medical Reference (QMR; version 3.7.1), were used by participants for diagnostic evaluation of a total of 36 cases based on actual patients. After training, each subject evaluated 9 of the 36 cases, first without and then using a DSS, and suggested an ordered list of diagnostic hypotheses after each evaluation. MAIN OUTCOME MEASURE: Diagnostic accuracy, measured as the presence of the correct diagnosis on the hypothesis list and also using a derived diagnostic quality score, before and after consultation with the DSSs. RESULTS: Correct diagnoses appeared in subjects' hypothesis lists for 39.5% of cases prior to consultation and 45.4% of cases after consultation. Subjects' mean diagnostic quality scores increased from 5.7 (95% confidence interval [CI], 5.5-5.9) to 6.1 (95% CI, 5.9-6.3) (effect size: Cohen d = 0.32; 95% CI, 0.23-0.41; P<.001). Larger increases (P = .048) were observed for students than for residents and faculty. Effect size varied significantly (P<.02) by DSS (Cohen d = 0.20; 95% CI, 0.08-0.32 for ILIAD vs Cohen d = 0.45; 95% CI, 0.31-0.59 for QMR). CONCLUSIONS: Our study supports the idea that "hands-on" use of diagnostic DSSs can influence diagnostic reasoning of clinicians. The larger effect for students suggests a possible educational role for these systems.

Patient or physician preferences for decision analysis: the prenatal genetic testing decision.

The choice between amniocentesis and chorionic villus sampling for prenatal genetic testing involves tradeoffs of the benefits and risks of the tests. Decision analysis is a method of explicitly weighing such tradeoffs. The authors examined the relationship between prenatal test choices made by patients and the choices prescribed by decision-analytic models based on their preferences, and separate models based on the preferences of their physicians. Preferences were assessed using written scenarios describing prenatal testing outcomes, and were recorded on linear rating scales. After adjustment for sociodemographic and obstetric confounders, test choice was significantly associated with the choice of decision models based on patient preferences (odds ratio 4.44; Cl, 2.53 to 7.78), but not with the choice of models based on the preferences of the physicians (odds ratio 1.60; Cl, 0.79 to 3.26). Agreement between decision analyses based on patient preferences and on physician preferences was little better than chance (kappa = 0.085+/-0.063). These results were robust both to changes in the decision-analytic probabilities and to changes in the model structure itself to simulate non-expected utility decision rules. The authors conclude that patient but not physician preferences, incorporated in decision models, correspond to the choice of amniocentesis or chorionic villus sampling made by the patient. Nevertheless, because patient preferences were assessed after referral for genetic testing, prospective preference-assessment studies will be necessary to confirm this association.

The role of physician preferences in the choice of amniocentesis or chorionic villus sampling for prenatal genetic testing.

Our objective was to determine the effect of physician preferences, as well as physician demographic, obstetric, and practice-related factors, on the choice of prenatal test made by their patients. We studied preferences for prenatal outcomes for 372 pregnant women who either chose amniocentesis (AMN) (n = 288) or chorionic villus sampling (CVS) (n = 84) for the indication of maternal age. We also studied preferences for these outcomes for the 92 physicians that referred them for testing. Preferences were assessed using written scenarios and were measured on linear rating scales. According to patients, the choice of prenatal test was made entirely or mostly by the physician in 14% of cases and was shared equally between patient and physician in 37% of cases. After adjustment for patient preferences, physician concern about spontaneous abortion of a normal fetus after CVS (odds ratio 0.71; CI, 0.48-1.05; p = 0.08), and a limb reduction (LRD) birth after CVS (odds ratio 0.85; CI, 0.68-1.05; p = 0.12), tended to decrease their patients' odds of choosing CVS, but the results were not statistically significant. No other physician preference, and no physician demographic, obstetric, or practice-related factor, influenced patient test choice. We conclude that after taking patient preferences into account, physician preferences and practice-related factors did not emerge as significant determinants of the choice of prenatal test made by their patients. It remains possible, however, that physician concern about spontaneous abortion and about LRD increase the likelihood of their patients choosing AMN over CVS.

Prenatal testing for limb reduction defects. How patients' views affect their choice of CVS.

OBJECTIVE: To determine the effect of reports and media coverage on chorionic villus sampling (CVS) and limb reduction defects (LRD) on patients' utilization of CVS for prenatal testing for advanced maternal age and to quantitate the relation of preferences for CVS and amniocentesis (AMN) outcomes to test utilization. STUDY DESIGN: We compared CVS and AMN utilization rates in two groups of women seen at one academic medical center before and after publicity concerning CVS and LRD. We measured preferences, in rating-scale units, for potential outcomes of prenatal testing in the after-publicity group. Relationships between preferences and CVS utilization were examined using multivariate methods. RESULTS: The proportion of women utilizing CVS for prenatal testing declined significantly for the after-publicity group (23%) as compared with the prepublicity group (47.4%, P = .0001). Belief that the birth of a child with LRD after CVS was worse than a similar birth after AMN significantly reduced the odds of choosing CVS. A decrement in preference for a child with LRD after CVS of 5 rating-scale units reduced the likelihood of CVS by 15% (confidence interval [CI], 1-28%); a decrement of 10 units reduced the likelihood of CVS by 28% (CI, 1-48%). This effect persisted after adjustment for demographic and obstetric covariates, prior prenatal testing, locus of decision making (patient versus physician or shared) and other maternal preferences for outcomes of prenatal testing. CONCLUSION: Lower patient preference for a child with LRD after CVS was an independent predictor of choosing AMN over CVS and was probably responsible for the significant decrease in CVS utilization at our hospital.

Bromide intoxication due to propantheline bromide.

An elderly woman presented with pneumonia and mental status changes and was found to have bromide poisoning due to ingestion of propantheline bromide over a 2-month period. The interference of bromide with serum chloride measurements on an ion-selective electrode resulted in spurious hyperchloremia and was crucial in making the diagnosis. To our knowledge, bromide intoxication due to propantheline bromide has not been reported previously.

Ethylene glycol poisoning with a normal anion gap caused by concurrent ethanol ingestion: importance of the osmolal gap.

Ethylene glycol poisoning classically presents as a metabolic acidosis with an increased anion gap. Metabolism of ethylene glycol to organic acids, and increased production of lactate, are responsible for the increased gap. We report the case of an alcohol user who consumed ethanol and ethylene glycol concurrently, and presented without acidosis, with a normal anion gap. Several hours later, when his serum ethanol level had declined, he developed severe acidosis with an elevated anion gap. An increased osmolal gap, not accounted for by the serum ethanol level, proved to be an important clue to the diagnosis. In this patient, ingestion of ethanol inhibited the hepatic metabolism of ethylene glycol to organic acids, obscuring the diagnosis. In intoxicated alcohol users, even in the absence of metabolic acidosis, serum osmolality measurements and calculation of the osmolal gap may facilitate the rapid diagnosis of ethylene glycol poisoning.

Use of decision analysis to evaluate patients' choices of diagnostic prenatal test.

Women with a family history of a chromosomal or genetic abnormality must weigh several factors in choosing between amniocentesis and chorionic villus sampling. We compared the prenatal test choices of three such women with those of decision analytic models that incorporated their preferences. Patient preferences were assessed using visual linear rating scales. Threshold analysis was used to determine preference ranges, and stochastic sensitivity analysis to provide confidence levels, for each choice of test. The test choices of patients and decision analytic models agreed in one case, and disagreed in two cases. In one of the latter two cases, stochastic and threshold analyses showed the disagreement to be slight; for small shifts in preference differences for first- vs. second-trimester diagnosis, or first- vs. second-trimester therapeutic abortion, patient and decision model would have agreed. In the other, stochastic analysis showed their differences to be large; there were no thresholds for early diagnosis, or for early therapeutic abortion, that would have led to agreement between patient and model. In the two cases in which patient and decision model agreed or slightly disagreed, the patients had made their own choice of prenatal test. In the case in which patient and decision model strongly disagreed, the patient's physician had shared in the choice of test. Decision analysis can be useful in analyzing prenatal test choices based on individual preferences for pregnancy outcomes. When choices of patients and decision models do not agree, examination of the locus of decision making (patient vs. physician) may help resolve apparent differences.

Communication with deaf patients. Knowledge, beliefs, and practices of physicians.

OBJECTIVE: To assess physicians' knowledge and beliefs regarding communication with deaf people and compare their knowledge and beliefs with their methods of communicating with deaf patients in their practices. DESIGN: Survey. SETTING: University medical center. SUBJECTS: Attending physicians in an internal medicine department. INTERVENTIONS: Physicians were surveyed regarding prior contacts with deaf patients and with deaf people outside the medical setting, and regarding their knowledge and beliefs concerning methods of communicating with deaf people. Physicians were asked to estimate the fraction of encounters in which they communicated with deaf patients by lipreading, writing, translation by a relative or friend, a sign language interpreter, or other methods. RESULTS: Writing was the method used most frequently in communicating with deaf patients. Although 63% of physicians knew that signing should be the initial method of communicating with deaf patients who sign, only 22% used sign language interpreters more frequently than other methods in their practices. Past contact with deaf people (P = .05), belief that communication by signing was the best means of communication (P = .04), and knowledge of the inefficiency of lipreading (P = .04) were predictors of the use of sign language interpreters for deaf patients. Physicians who used sign language interpreters more frequently than other methods believed that much more time and effort were involved in caring for deaf than for hearing patients compared with those who used interpreters less frequently (P = .08). CONCLUSION: Although most physicians believed that use of sign language interpreters was preferable, only a minority used them in their practices. Greater recognition of the advantages of signing over other methods and greater availability of sign language interpreters should lead to more effective communication between deaf patients and physicians.

Preferences of pregnant women for amniocentesis or chorionic villus sampling for prenatal testing: comparison of patients' choices and those of a decision-analytic model.

Decision analytic models have suggested that the choice of amniocentesis or chorionic villus sampling for prenatal genetic testing is a utility-driven decision. We compared preferences for prenatal testing among 156 pregnant women who had chosen either amniocentesis (n = 82) or chorionic villus sampling (n = 74) for the indication of maternal age. We also compared their choices with those of a decision-analytic model based on their preferences, and age-specific rates of spontaneous abortion and chromosomal abnormalities. Preferences were assessed using written scenarios describing potential outcomes of prenatal testing, and were recorded on linear rating scales. The differences in preference ratings for first- vs second-trimester prenatal diagnosis of a normal child (4.2 vs -1.6, p = 0.0004), and for first- vs second-trimester abortion of an abnormal fetus (4.4 vs -1.6, p = 0.01), were significantly greater among women choosing chorionic villus sampling than among women choosing amniocentesis. There were no significant differences between chorionic villus sampling and amniocentesis patients in their preference ratings for test-related miscarriage, disconfirmed results at pregnancy termination, or maternal morbidity from therapeutic abortion. After adjusting for demographic and obstetric factors, the difference in preferences for early vs late prenatal diagnosis was an independent predictor of the choice of chorionic villus sampling in a multivariate model. Among women whose decision analyses selected amniocentesis, 56.8% had chosen amniocentesis, and among women whose analyses selected chorionic villus sampling, 63.2% had chosen chorionic villus sampling (p = 0.05). We conclude that the preferences of pregnant women for the outcomes of prenatal testing were associated with their choice of amniocentesis or chorionic villus sampling. In addition, the choice of prenatal test made by the majority of women was concordant with that of a decision-analytic model that incorporated their preferences. Nevertheless, because many women made choices that were discordant with their decision-analytic results, further research into the bases for their choices is warranted.

A cost-effectiveness analysis of amniocentesis and chorionic villus sampling for prenatal genetic testing.

Amniocentesis and chorionic villus sampling are tests for the prenatal diagnosis of cytogenetic abnormalities. We calculated the incremental costs per abnormal birth averted, and the incremental costs per quality-adjusted outcome, of amniocentesis and chorionic villus sampling performed for the indication of maternal age. Probabilities were obtained from the literature, and direct medical costs from hospital charges deflated to reflect aggregated contracted care reimbursements. Utilities were used to quality-adjust prenatal testing outcomes. Based on costs per abnormal birth averted, at all maternal ages from 30 to 43 years, amniocentesis was more cost-effective than chorionic villus sampling; at ages 44 and 45, chorionic villus sampling was more cost-effective. However, if the anxiety reduction provided by first-trimester diagnosis was equivalent to a 0.2% risk of an abnormal child, chorionic villus sampling was more cost-effective than amniocentesis at all maternal ages. Based on data from the 1988 United States natality cohort, the current policy of testing women aged 35 and older would cost $103,329 and $111,184 per abnormal birth averted for amniocentesis and chorionic villus sampling, respectively. Testing women aged 30 and older would almost double these costs. For either prenatal test, targeting high risk women for testing, and striving for utilization rates of 50% or higher, appeared to be the most cost-effective policy.

Accuracy and reproducibility of precordial percussion and palpation for detecting increased left ventricular end-diastolic volume and mass. A comparison of physical findings and ultrafast computed tomography of the heart.

OBJECTIVE: To assess the accuracy and reproducibility of indirect definitive precordial percussion in detecting increased left ventricular end-diastolic volume (LVEDV), left ventricular mass (LVM), and left ventricular end-diastolic wall thickness (LVEDWT), and to compare it with palpation of the apical impulse. DESIGN: Descriptive study. SETTING: Hospitals and clinics of a university medical center. PATIENTS: Convenience sample of 103 patients (62 men and 41 women) referred for ultrafast computed tomography (CT) of the heart. INTERVENTIONS: Percussion dullness distance from the midsternal line in the left fourth through sixth intercostal spaces, distance of the apical impulse from the midsternal line, and apical impulse diameter in the left lateral decubitus position were measured on all patients. Measurements of LVEDV, LVM, and LVEDWT were taken using ultrafast CT of the heart. Investigators performing the physical diagnostic maneuvers were blinded to the clinical history and CT results, and investigators performing the CT scans were blinded to physical findings. RESULTS: Percussion dullness distance in the left fifth intercostal space was the best discriminator of LVEDV (receiver operating characteristic [ROC] area, 0.680; 95% confidence interval [CI], 0.547 to 0.813), and dullness distance in the left sixth intercostal space was the best discriminator of LVM and LVEDWT (ROC areas, 0.831, 95% CI, 0.674 to 0.988; and 0.849, 95% CI, 0.651 to 0.999, respectively). A percussion dullness distance of greater than 10.5 cm in the left fifth intercostal space detected increased LVEDV or LVM with a sensitivity of 91.3% (95% CI, 70.5% to 98.5%) and a specificity of 30.3% (95% CI, 19.9% to 43.0%). There was moderate concordance between investigators for percussion dullness distance (kappa, 0.57; 95% CI, 0.18 to 0.96). In patients in whom an impulse was palpated, an apical impulse diameter of greater than 3.0 cm in the left lateral decubitus detected increased LVEDV or LVM with a sensitivity of 100% (95% CI, 77.1% to 100%) and a specificity of 40% (95% CI, 23.2% to 59.3%). However, an impulse was palpable in only 53% of cases and showed only slight interobserver reproducibility (kappa, 0.18; 95% CI, 0.0 to 0.58). CONCLUSION: Indirect definitive percussion of the precordium is a sensitive and moderately reproducible maneuver for excluding cardiomegaly due to increased LVEDV or LVM. Although measurement of apical impulse diameter was also sensitive in excluding cardiomegaly, lack of a palpable impulse in many patients and low precision between physicians may limit its utility in clinical practice.

Discrimination and reproducibility of an information maximizing multivariable model.

Predictor variables for multivariate rules are frequently selected by methods that maximize likelihood rather than information. We compared the discrimination and reproducibility of a prediction rule for pneumonia derived using extended dependency analysis (EDA), an information maximizing variable selection program, with that of a validated rule derived using logistic regression. Discrimination was measured by receiver-operating characteristic (ROC) analysis, and reproducibility by rederivation of the rule on 200 replicate samples of size 250 and 500, generated from a training cohort of 905 patients using Monte Carlo techniques. Four of the five predictor variables selected by EDA were identical to those selected by logistic regression. With each variable weighted by its conditional contribution to total information transmission, EDA discriminated pneumonia and nonpneumonia in the training cohort with an ROC area of 0.800 (vs 0.816 for logistic regression, p = 0.60), and in the validation cohort with an area of 0.822 (vs 0.821 for logistic regression, p = 0.98). EDA demonstrated reproducibility comparable to that of logistic regression according to most criteria for replicability. Replicate EDA models showed good discrimination in the training and testing cohorts, and met statistical criteria for validation (no significant difference in ROC areas at a one-tailed alpha level of 0.05) in 80.8% to 94.2% of cases. We conclude that extended dependency analysis selected the most important variables for predicting pneumonia, based on a validated logistic regression model. The information-theoretic model showed good discriminatory power, and demonstrated reproducibility according to clinically reasonable criteria. Information-theoretic variable selection by extended dependency analysis appears to be a reasonable basis for developing clinical prediction rules.

Diagnostic reasoning of high- and low-domain-knowledge clinicians: a reanalysis.

Thinking-aloud protocols provided by Joseph and Patel were reanalyzed to determine the extent to which their conclusions could be replicated by independently developed coding schemes. The data set consisted of protocols from four cardiologists (low domain knowledge = LDK) and four endocrinologists (high domain knowledge = HDK), individually working on a diagnostic problem in endocrinology. The two analyses agree that the HDK physicians related data to potential diagnoses more than did the LDK group and were more focused on the correct diagnostic components. However, the reanalysis found no meaningful difference between the groups in diagnostic accuracy, speed of diagnosis, or the breadth of the search space used to seek a solution. In the reanalysis, the HDK physicians employed more single-cue inference and less multiple-cue inference. The generalizability of results of protocol-analysis studies can be assessed by using several complementary coding schemes.

Attitudes of funeral directors and embalmers toward autopsy.

Although physician and family attitudes toward autopsy have been suggested as factors leading to declining autopsy rates, attitudes of funeral directors and embalmers toward autopsy have not been studied. We surveyed members of the Illinois Funeral Directors Association concerning beliefs about the purposes of autopsy, problems they experienced with autopsied cases, and the advice they gave to family members about permitting autopsy. Three hundred eight (42.2%) of 730 funeral directors and embalmers responded to the questionnaire. Although 80.3% believed that autopsy served a purpose, 46.4% had counseled families not to permit autopsy, and 16.6% did so more than half the time. Perceived difficulty in embalming autopsied cases (odds ratio, 2.3), family concern about the risk of disfigurement (odds ratio, 1.9 to 2.2), having attempted to talk to a pathologist or hospital administrator about a poorly performed autopsy (odds ratio, 2.0), and belief in the purpose of autopsy (odds ratio, 0.38), were significant independent predictors of counseling against autopsy. Of funeral directors who counseled against autopsy, 28.3% reported that families never permitted autopsy, and 39.4% reported that families only occasionally permitted autopsy, in cases where they had counseled against it. Funeral directors from Illinois counties with 1989 medical certificate case autopsy rates of less than 5% counseled against autopsy more frequently than funeral directors from counties with higher autopsy rates. We conclude that although most funeral directors and embalmers believe in the purposes of the autopsy, difficulties in embalming autopsied cases, and family concern about the risk of disfigurement, may lead members of the funeral profession to counsel against autopsy, and that such advice may influence families not to permit an autopsy.

Reproducibility of predictor variables from a validated clinical rule.

It has been suggested that clinical prediction rules are not reproducible, and that the most important variables frequently do not appear in replicate models. The authors studied the reproducibility of a validated rule for predicting radiographic evidence of pneumonia (ROC areas for the training and validation cohorts, 0.816 and 0.821, respectively). Two hundred replicate samples of size 250 and size 500 were generated by sampling without replacement from the original training cohort of 905 patients with a 14.6% prevalence of pneumonia. Forward selection was performed among 31 candidate variables by stepwise logistic regression. Using as reproducibility criteria: 1) inclusion of all five variables from the original model in the original order; 2) inclusion of all five variables in any order; 3) inclusion of the first three variables; 4) inclusion of the first two variables; 5) inclusion of the first variable; and 6) inclusion of any of the five variables: 2.5%, 13.5%, 48.5%, 85.5%, 98.0%, and 100% of replicate models of sample size 500, respectively, met the criteria, whereas 0%, 0%, 16.5%, 49.0%, 71.5%, and 97.5% of models of sample size 250 met the criteria (all comparisons by sample size p < .0001 except for criteria 1 and 6, p = 0.07). Mean ROC areas in the training and validation samples were 0.829 and 0.791 for replicate models of sample size 500, and 0.831 and 0.779 for models of sample size 250. There was no significant difference in ROC areas between training and validation cohorts for 80.5% of models of sample size 500, and for 75.3% of models of sample size 250.(ABSTRACT TRUNCATED AT 250 WORDS)