Search

To determine the relative value of carotid phonoangiography (CPA), oculoplethysmography-Kartchner (OPG-K), and Doppler ultrasonic arteriography (UA), 90 vessels undergoing carotid endarterectomy were prospectively examined. By analyzing the data on receiver operator characteristic curves, the dynamic relationship between sensitivity and specificity for each of the three noninvasive tests was assessed. Disease was defined by either the percentage of angiographic stenosis or the mean pressure gradient across the carotid (deltaP). All three tests were shown to be relatively insensitive, but quite specific, if disease was defined by 50% and 60% angiographic stenosis or deltaP of greater than 10 and 20 mm Hg. By employing a more rigid definition of disease, 70% stenosis or deltaP of greater than 30 mm Hg, sensitivity was increased for all examinations and was highest in OPG-K and UA for a given specificity. The sensitivity for UA was enhanced to 80% with a comparable specificity, if those 23 UA exams with plaque were treated as positive studies. The combination of CPA, OPG-K, and UA was superior to any one of these tests alone, but the best value balancing maximum sensitivity and specificity still was associated with a 23% false negative rate. This study would suggest that these three tests should be limited to screening patients at risk for carotid stenosis and not for symptomatic patients. To achieve the best balance between sensitivity and specificity, lax threshold criteria for calling the test positive should be employed, and the tests should be used in combination.

The administration of digitalis by experienced cardiologists has been examined and a prototypical computer program has been developed which captures portions of their expertise. The new program first constructs a patient-specific model upon which to base the determination of dosage and then uses feedback information about a variety of clinical aspects of the patient's response to modify its recommendations. The model reflects both the program's knowledge of pharmacokinetics and those special features of the patient's condition which may alter his response to therapy. The program makes assessments of the therapeutic and the toxic effects of digitalis on the patient, and bases its subsequent recommendations on the "therapeutic-toxic" state which best describes the evolving clinical situation. A clinical trial was performed in which the program "followed" a series of patients managed by clinicians on a cardiology service. That trial demonstrated the feasibility of this type of program in dealing with acutely ill patients, even those who have increased sensitivity to the toxic effects of digitalis. Each patient in the trial in whom toxicity developed had received more digitalis than would have been recommended by the program. This approach to automated clinical consultation should eventually provide a technology for the distribution of clinical expertise.

Patients with polycystic kidney disease are at increased risk of subarachnoid hemorrhage from rupture of intracranial aneurysms. We used decision analysis to assess whether or not patients with polycystic kidney disease should undergo routine cerebral arteriography for intracranial aneurysms and prophylactic surgery, if an aneurysm is detected. We incorporated published data on the prevalence of intracranial aneurysms in patients with polycystic kidney disease, the annual rate of aneurysmal rupture, the risk of grave complications of rupture, and the likelihood of grave complications of arteriography and prophylactic surgery. Outcomes were assessed as years of survival, and benefit was calculated as the gain in survival. Our analysis shows that arteriography should not be carried out routinely because its benefit exceeds one year only if the prevalence of aneurysm exceeds 30 per cent, if the surgical complication rate is 1 per cent or less, and if the patient is under 25 years of age. If newer noninvasive tests, such as digital-subtraction angiography, prove to identify reliably patients who are highly likely to have a cerebral aneurysm, routine screening with these tests will be warranted in patients with polycystic kidney disease.

The quantitative use of patients' attitudes in medicine has thus far been limited to decisions involving either treatment alternatives or the use or nonuse of a particular diagnostic test. Preference theory has not been applied either to the use of screening tests or to the development of large-scale health-related public policy decisions. In this paper we have, in a prototypical fashion, analyzed the effect patient attitudes have on a public policy decision faced by many countries today--whether or not to institute a screening program for neural tube defects. We have assessed the attitudes of 338 prospective parents toward many of the sequelae expected from the introduction, or lack thereof, of the alpha-fetoprotein screening program--induced abortion from amniocentesis, elective abortion, and the birth of a defective child. Using these data and information collected by the United Kingdom study on alpha-fetoprotein, we have estimated the proportion of patients coming to genetic counseling who would benefit from the availability of a screening program for neural tube defects.

Since 1978, decision analysis consultations have been offered to help physicians manage particularly vexing clinical problems. Consultations were requested for decisions perceived as difficult because: the diagnosis was uncertain; the available therapy had uncertain efficacy or risks; the patient had a short life expectancy that limited the potential benefit of therapy; the risk of a given test or therapy was increased and the usual rules for applying the test or giving the treatment did not apply; the need for a therapeutic procedure was acknowledged, but competing risks created uncertainty about the optimal timing of the procedure; the optimal sequence of multiple necessary procedures could not be discerned readily; explicit patient input into decision-making was required; certain medical information could not be interpreted easily; and a rare, unique, or new problem was encountered. To carry out these consultations, standard techniques were used, including decision tree models, Bayes' theorem, Markov analysis, and utility assessment, but old approaches were modified to adapt decision analysis to individual patient problems. This experience demonstrates that decision analysis can be carried out effectively on a consultative basis for individual patients. A consultation service can also train analysts in decision-making and drive research in medical problem solving.

Variability in length of stay (LOS) and its determinants were studied in 60 patients without complications after acute myocardial infarction (AMI) at three teaching hospitals. The patients at the three hospitals had similar clinical presentations and in-hospital courses. However, 81% of patients admitted to the intensive care unit (ICU) at one hospital were transferred to lower levels of care within 48 hours, compared with 40% and 56% at the other two hospitals (p less than 0.05). Patients who underwent pre-discharge exercise tests (ETT) had a longer total LOS than patients who did not (11.9 +/- 4.3 vs 9.2 +/- 2.9 days; p less than 0.01). Similarly, patients who underwent 24-hour ambulatory monitoring (AMB-MON) had a longer total LOS than patients who did not (12.3 +/- 4.3 vs 9.1 +/- 2.8 days; p less than 0.001). No differences in clinical characteristics or in survival at six months were detected between patients who did and did not undergo ETT or AMB-MON. Total LOS correlated with whether patients underwent 0, 1, or both of these tests (p less than 0.0001); each test was associated with a 2.3-day increase in the duration of hospitalization. The authors conclude that such tests may prolong total LOS of patients without complications after AMI.

Over the past decade we have used a decision analytic model to counsel 840 patients (468 women, 381 men, 432 couples) about amniocentesis for prenatal diagnosis. The model explicitly considers the possibilities of miscarriage (both after amniocentesis and spontaneously), an affected child (as a function of maternal age), and various diagnostic errors. Prospective parents are shown the model after routine counseling is used to explain the options and potential outcomes. Using the lottery technique (in which they are asked to choose between therapeutic abortion and carrying a pregnancy to term without the benefit of amniocentesis, where the likelihood of an affected child is varied in a structured sequence), prospective parents expressed their attitudes on a utility scale, where zero corresponds to an unaffected child and where 100 corresponds to an affected child. On that scale, the mean assessed disutility of therapeutic abortion was 33.7 +/- 32.6 (35.8 +/- 32.1 among women, 30.9 +/- 32.9 among men). The decision model encourages couples to confront their attitudes toward specific reproductive outcomes, to clarify their values and to incorporate them, along with their current risks, into a logical decision about prenatal diagnosis.

The authors developed a decision tree-critiquing program (called BUNYAN) that identifies potential modeling errors in medical decision trees. The program's critiques are based on the structure of a decision problem, obtained from an abstract description specifying only the basic semantic categories of the model's components. A taxonomy of node and branch types supplies the primitive building blocks for representing decision trees. Bunyan detects potential problems in a model by matching general pattern expressions that refer to these primitives. A small set of general principles justifies critiquing rules that detect four categories of potential structural problems: impossible strategies, dominated strategies, unaccountable violations of symmetry, and omission of apparently reasonable strategies. Although critiquing based on structure alone has clear limitations, principled structural analysis constitutes the core of a methodology for reasoning about decision models.

Patients with lung cancer can be treated by either surgical extirpation or radiation. The former may offer increased five-year survival and prolonged life expectancies as compared to the latter, but subjects patients to the immediate risk of thoracotomy. We interviewed patients with "operable" lung cancer and found that they were quite averse to taking risks involving the possibility of immediate death. When these data about patients' attitudes were combined with data about survival after both radiation therapy and operation, it appeared that radiotherapy would be the preferred therapeutic plan for several of these patients. These results emphasize the importance of choosing therapies not only on the basis of objective measures of survival but also on the basis of patient attitudes.

Many difficult medical decisions involve uncertainty. Decision analysis-an explicit, normative and analytic approach to making decisions under uncertainty-provides a probabilistic framework for exploring difficult problems in nondeterministic domains. As the methodology has advanced, clinical decision analysis has been applied to increasingly complex medical problems and disseminated widely in the medical literature. Unfortunately, this approach imposes a heavy computational burden on analysts. Microcomputer-based decision-support software can ease this burden.

To help the physician decide whether or not to treat a patient who may or may not have a disease, a method has been developed for calculating a therapeutic threshold. If the probability of disease in a given patient exceeds the threshold, the preferable course of action is to treat; if the probability is below the threshold, the preferable course of action is to withhold treatment. This method is applicable in many medical and surgical settings in which some diagnostic uncertainty exists after all appropriate studies have been carried out. The technic not only exposes some of the basic principles of therapeutic decision making in the face of diagnostic uncertainty but also forms a convenient framework for analyzing the impact of "soft" clinical data on the decision-making process.

Remarkably little is known about the cognitive processes which are employed in the solution of clinical problems. This paucity of information is probably accounted for in large part by the lack of suitable analytic tools for the study of the physician's thought processes. Here we report on the use of the computer as a laboratory for the study of clinical cognition. Our experimental approach has consisted of several elements. First, cognitive insights gained from the study of clinicians' behavior were used to develop a computer program designed to take the present illness of a patient with edema. The program was then tested with a series of prototypical cases, and the present illnesses generated by the computer were compared to those taken by the clinicians in our group. Discrepant behavior on the part of the program was taken as a stimulus for further refinement of the evolving cognitive theory of the present illness. Corresponding refinements were made in the program, and the process of testing and revision was continued until the program's behavior closely resembled that of the clinicians. The advances in computer science that made this effort possible include "goal-directed" programming, pattern-matching and a large associative memory, all of which are products of research in the field known as "artificial intelligence". The information used by the program is organized in a highly connected set of associations which is used to guide such activities as checking the validity of facts, generating and testing hypotheses, and constructing a coherent picture of the patient. As the program pursues its interrelated goals of information gathering and diagnosis, it uses knowledge of diseases and pathophysiology, as well as "common sense", to dynamically assemble many small problem-solving strategies into an integrated history-taking process. We suggest that the present experimental approach will facilitate accomplishment of the long-term goal of disseminating clinical expertise via the computer.

The physician developing a treatment plan for a particular patient often needs to know the life expectancy associated with the outcomes of therapeutic choices. Currently available methods for estimating life expectancy are cumbersome and of limited clinical use. We describe a simple approximation of life expectancy (the "DEALE") that is based on the assumption that survival follows a simple declining exponential function. In this approach, the reciprocal of the age-, sex-, and race-adjusted life expectancy is used to estimate the mortality rate of a healthy person. The life expectancy of a person who also has one or more diseases is obtained by adding disease-specific mortalities to the age-, sex-, and race-adjusted mortality rate and taking the reciprocal of that sum. In this paper we show that this approximation estimates life expectancy accurately for the great majority of clinical problems.

Since its introduction into medicine 15 years ago, decision analysis has been applied to difficult clinical problems. Several important advances have made the process more practical and acceptable: computer programs that eliminate the need for burdensome calculations, improved techniques for designing analytic models, the ability to carry out sensitivity analyses over several dimensions simultaneously, and the elaboration of clinically relevant measures of utility. Using these techniques, analysts have addressed many important clinical issues including screening for and prevention of disease, tradeoffs among tests and treatments, and the interpretation of clinical data under conditions of uncertainty. Problems with the approach remain and applications have not been extensive, but decision analysis is evolving as a powerful clinical tool and gradually is gaining acceptance in medical practice.

DT, a 63-year-old white male with insulin-dependent diabetes mellitus and severe peripheral vascular disease, was admitted with a five-day history of vague abdominal pain and diarrhea. On the day of admission he vomited three times, was noted to have a bloody stool, and came to the emergency room. DT denied hematemesis, fever, or chills. He had bilateral leg amputations and had sustained three myocardial infarctions, the last one 15 months before this admission. He had never experienced symptoms of abdominal angina. Of significance was his history of congestive heart failure, mitral regurgitation, and atrial fibrillation. His medications on admission included digoxin 0.25mg per day, furosemide 40mg per day, and NPH insulin 15 units per day. On admission to the hospital his oral temperature was 38 degrees C, pulse was 90/min, respiratory rate was 24/min, and blood pressure was 134/80mmHg. Abdominal examination revealed a distended abdomen with hypoactive bowel sounds and mild tenderness. Chest x ray revealed cardiomegaly. The electrocardiogram demonstrated atrial fibrillation. A plain film of the abdomen was positive for gallstones and edema of the bowel wall (thumb-printing). Laboratory results included blood urea nitrogen 48mg%, creatinine 1.2mg%, hemoglobin 18g/dl, and hematocrit 52.9%. White blood cell count was 11,900 cells/cc with 33% polymorphonuclear leukocytes, 47% bands, 8% lymphocytes, 11% monocytes, and 1% atypical lymphocytes. The prime considerations for differential diagnosis were mesenteric ischemia and infectious gastroenteritis. While it was appreciated that mesenteric ischemia, if present, might warrant surgical intervention, the risk of anesthesia itself in this patient was felt by his attending physicians to exceed 30%. Furthermore, the clinical findings were only "suggestive" of mesenteric eschemia. They were certainly not "diagnostic." In view of this dilemma, a consultation with the Division of Clinical Decision Making was requested.

In a registry of volunteer bone marrow donors, the relation between registry size and probability of finding an exact or partial match for a random recipient cannot be theoretically derived because it depends on specifics of the human leukocyte antigen (HLA) haplotype frequencies in the donor and recipient populations. The relation must be explicitly calculated using empirically determined HLA haplotype frequency data for all possible pairings between a donor and a recipient population. This report describes a general solution to this problem. The method shows that the relation of the probability of matching to registry size is sigmoidal, with small increases in probability at the extremes of registry size and a middle range of registry size within which the probability of matching increases most sharply. This range determines the approximate size of the most cost-effective registry. In addition, for any pairing of donor and recipient populations, there is a maximum probability of identifying a match of a given quality for a random recipient, which cannot be exceeded even if registry size were infinite. This upper limit is a function of the frequency of blank (or unknown) alleles in the donor and recipient populations; the higher that frequency, the lower the maximum probability of achieving any given quality of match. The determinants of the probability of achieving a given quality of match with a given registry size are (1) the genetic heterogeneity within the recipient and donor populations, which increases the registry size required to achieve a given probability of matching, and (2) the degree of genetic homology between the donor and recipient populations, which increases the maximum probability of matching and also lowers registry size requirements. The method described here can be used to estimate donor pool size requirements using any donor and recipient populations for which HLA frequency data are available.

Pathologists typically analyze biopsy specimens and report their opinions in a qualitative fashion. Clinical information, often fragmentary in character, is implicitly linked with the histologic findings, and interpretations are couched in ambiguous language: "consistent with," "highly suggestive of," "may represent," or "cannot exclude." Only pathognomonic or normal findings are reported in an unequivocal fashion. In this paper we compare the analysis of biopsy material by a conventional method and by a numerical, probabilistic technique. We suggest that applying a probabilistic technique can considerably improve the precision of tissue diagnosis and can greatly facilitate the communication of pathologists with clinicians and with each other. Probabilistic analysis is also likely to be of substantial value in improving the interpretation and reporting of x-ray and nuclear-medicine studies.

The physician's estimate of prognosis under alternative treatment plans is a principal factor in therapeutic decision making. Current methods of reporting prognosis, which include five-year survivals, survival curves, and quality-adjusted life expectancy, are crude estimates of natural history. In this paper we describe a general-purpose model of medical prognosis based on the Markov process and show how this simple mathematical tool may be used to generate detailed and accurate assessments of life expectancy and health status.

The management of the adult tuberculin reactor who has no other risk factors for the development of active tuberculosis is controversial. We performed a decision analysis to address the issue of whether adults 20 to 80 yr of age should take isoniazid to prevent the development of active disease. The analysis uses a Markov simulation, a regression model to calculate the risk of developing tuberculosis at any given time after the discovery of a positive skin test, and estimates from the literature for other probabilities. The analysis favors withholding INH for all age groups, but by only 4 to 17 days of life expectancy, depending on the age of the patient. Quality adjusting and discounting change neither the preferred strategy nor the closeness of the outcome. Sensitivity analyses also underscore the narrowness of the margin. Our analysis reaches different conclusions from previously published analyses. More important than the actual strategy preferred is the minute difference in expected utilities between administering and withholding INH. We suggest that the individual patient decide whether or not these differences are meaningful.

We designed an intricate model of the process of glaucoma screening, diagnosis, and treatment in order to evaluate the medical care costs of such efforts in relation to the benefits in terms of quality-adjusted years of vision saved, utilizing the economic principles of cost-effectiveness analysis. Although a relatively limited data base and numerous assumptions concerning the accuracy of diagnostic tests, the natural history of ocular hypertension and glaucoma, and the effectiveness of available treatment modalities, limit our ability to draw definitive conclusions concerning the cost-effectiveness of various glaucoma screening options, our analysis indicates that glaucoma screening is probably cost-effective when targeted at certain subgroups of the population. Our analysis suggests that changes in several aspects of existing screening policies may be appropriate if cost-effectiveness is to be used as one of the criteria for the efficient allocation of resources to and within screening programs. Specifically, combinations of screening tests and screening targeted at high risk populations such as blacks, diabetics, and relatives of glaucoma patients are probably more cost-effective than screening of the general population with a single test. In younger populations, the importance of detecting ocular hypertension argues for the use of tonometry. Ophthalmoscopy may be more cost-effective in older age groups in whom the higher prevalence of glaucoma outweighs the need for identifying ocular hypertensives. In the very elderly, automated perimetry becomes cost-effective since the detection of established field loss will have greater yield. In addition, it is evident that diversion of resources away from actual screening efforts and towards efforts aimed at improving follow-up and compliance would be an additional cost-effective strategy.

Formal prescriptive models can help patients and clinicians better understand the risks and uncertainties they face and better formulate well-reasoned decisions. Using Bayes rule, the clinician can interpret pedigrees, historical data, physical findings and laboratory data, providing individualized probabilities of various diagnoses and outcomes of pregnancy. With the advent of screening programs for genetic disease, it becomes increasingly important to consider the prior probabilities of disease when interpreting an abnormal screening test result. Decision trees provide a convenient formalism for structuring diagnostic, therapeutic and reproductive decisions; such trees can also enhance communication between clinicians and patients. Utility theory provides a mechanism for patients to understand the choices they face and to communicate their attitudes about potential reproductive outcomes in a manner which encourages the integration of those attitudes into appropriate decisions. Using a decision tree, the relevant probabilities and the patients' utilities, physicians can estimate the relative worth of various medical and reproductive options by calculating the expected utility of each. By performing relevant sensitivity analyses, clinicians and patients can understand the impact of various soft data, including the patients' attitudes toward various health outcomes, on the decision making process. Formal clinical decision analytic models can provide deeper understanding and improved decision making in clinical genetics.

When a diagnostic test is reported as normal, the clinician generally uses it only to rule out certain diseases. However, if properly interpreted, the normal value may help to differentiate among diagnoses that yield normal results with different frequencies. A simple method permits the extraction of such information. The physician estimates the probability of various diagnoses and then combines these estimates with the anticipated frequency of negative results for each disease under consideration.