Alvan R. Feinstein

The importance of classifying initial co-morbidity in evaluating the outcome of diabetes mellitus

Abstract The associated co-morbid diseases present at the time of discovery of adult maturity-onset diabetes mellitus have not hitherto been quantitatively related to the subsequent course of the patients. From existing medical records and other solicited data, complete 5-yr follow-up was obtained for an ‘inception cohort’ of 188 patients who were treated during the years 1959–1962 at the West Haven VA Hospital and whose diabetes was first diagnosed within 6 months before ‘zero time’, which was the date of discharge for the ailment that had provoked hospitalization. A special classification was developed for categorizing the zero-time co-morbidity of the patients as prognostically cogent or non-cogent, and for dividing cogent co-morbidity into vascular and nonvascular types, with functional severity graded as none, moderate, and severe. The fatality rate at 5 yr after zero time was 40 per cent (76/188) for all patients, but distinctive gradients of fatality were noted for age, for type of co-morbidity, and particularly for severity of co-morbidity. The 5-yr fatality rate after zero time was 7 per cent in forty-one patients with no cogent co-morbidity, 33 per cent in seventynine patients with moderate co-morbidity and 69 per cent in sixty-eight patients with severe co-morbidity. Of the sixty-eight patients with severe initial co-morbidity, 53 per cent later died of the same or a related ailment; in the seventy-nine patients with moderate co-morbidity, the analogous death rate was 13 per cent. Death was due to ‘vascular causes’ in 52 per cent of seventy-seven patients who initially had vascular co-morbidity, in 7 per cent of seventy patients with nonvascular co-morbidity, and in 2 per cent of forty-one patients with no cogent co-morbidity. Among 5-yr survivors, the occurrence rate of new vascular events (or ‘diabetic complications’) was directly related to the same features of age and co-morbidity that seemed to affect fatality. These data indicate that the outcome of patients with maturity-onset diabetes mellitus depends on the type and functional severity of the co-morbid diseases present when the diabetes is detected. An appropriate analysis of co-morbidity, although omitted from existing statistical studies of diabetes, is pre-requisite to evaluating the results of different modes of therapy.

Fraud, distortion, delusion, and consensus: The problems of human and natural deception in epidemiologic science

From the Clinical Epidemiology Unit, Yale University School of Medicine, New Haven, Connecticut, and Cooperative Studies Program Coordinating Center, Veterans Administration Medical Center, West Haven, Connecticut. This work was supported in part by a grant from the Andrew W. Mellon Foundation. Requests for reprints should be addressed to Dr. Alvan R. Feinstein, Yale University School of Medicine, I 456 SHM, P.O. Box 3333, New Haven, Connecticut 06510. Manuscript submitted November 25, 1987, and accepted December 15, 1987. During the past two decades, the quest for scientific truth in the world of medicine has had prominent problems when evidence was found to be fabricated, or when non-fabricated evidence evoked major controversies about its validity or interpretation. If the search for truth is both a professional and moral objective in the practice of scientific research, any deceptions that disguise truth are particularly repugnant. The degree of repugnance associated with different types of deception, however, will depend on the manner in which they are produced. Certain deceptions are deliberately planned frauds, but many others occur inadvertently. They arise as distortions or delusions, produced by flaws in the conventional methods used for the “standard” practice of scientific research. Although everyone is greatly distressed by the immoral aspects of fraud, the other types of deception have no associated ethical opprobrium. They are usually accepted as part of the inevitable risks associated with the benefits of scientific investigation. In this view, the research process is somewhat like an efficacious treatment that may occasionally have adverse side effects. We are pleased with the good things that occur in research; and we hope that the inevitable bad things will be infrequent or at least not too bad. Accordingly, vigorous counteractions are taken when deceptions are found to have been deliberate, but not when they are inadvertent. If an investigator paints mice, forges data, or creates non-existent patients, the scientific “immorality” is egregiously shocking and corrective measures take place promptly. People lose their jobs or reputations; old committees are reprimanded and new ones created; better procedures are developed for supervision, mentoring, and monitoring of research activities. After all the furor subsides, however, the research “industry” usually returns to its previous equilibrium. We feel better about the improvements instituted to avoid deliberate deceptions, but we give little or no attention to the much greater and more frequent problems of the inadvertent deceptions that occur when “established” standards of practice lead to distortions or delusions. In the inadvertent deceptions, no data or observations have been faked. The investigators may have been honest, dedicated, sincere, and hardworking; and the data may have been assembled and analyzed with conscientious care; but something went wrong. In a disfortion, something has affected the observed evidence submitted as “news”: the data may be inaccurate or the comparisons biased. In a delusion, something has affected the “editorials” that emerge when the observed evidence is interpreted to form hypotheses or conclusions. A distortion is usually produced by failure to recognize important distinctions in the complexity

A Critique of Methods in Reported Studies of Long-term Vascular Complications in Patients with Diabetes Mellitus

From 149 papers that referred to the development of vascular complications in patients with diabetes mellitus, we selected thirty-three that could be used for analysis of methodology. Eight of these reports contained inception cohorts in which a valid statistical risk could be calculated from the target events (cited in the numerators) and the initial states (cited in the denominators). Twenty-five reports contained longitudinal prevalence groups that were suitable only for appraisal of the target events cited in the numerators. The denominator populations in the cohort studies were usually reported without reproducible specifications of the proportionate sources of the cohort, the criteria for diagnosing diabetes mellitus, and the criteria for excluding patients from the investigation. Although most of the starting patients were “recovered” in reports of final results, the denominators of different results were often unstable, reflecting an absence of suitable initial data. The initial state of the cohorts was usually reported for age and sex, less commonly for metabolic features of the diabetes, and rarely for comorbid diseases that were unrelated to diabetes. The initial-state retinopathy, nephropathy, hypertension, peripheral vascular and cardiac ailments that were “related” to diabetes were reported infrequently. When reported, these conditions were not always specified with reproducible diagnostic criteria or cited quantitatively for prevalence in the cohort. The incidence of subsequent vascular complications was usually reported without satisfactory distinctions between the occurrence of a new complication and a change in the state of a previous one. The complications were generally cited with “objective” data that did not indicate the concomitant symptomatic and other functional effects. Few of the cohort investigators took appropriate advantage of the opportunity to correlate subsequent complications with different “risk factors” of the initial state. In longitudinal prevalence studies, the complications were correlated mainly with duration and regulation of the diabetes, but seldom with the initial distinctions of the patients. Since these problems arise from defects in selecting, specifying, and analyzing the original and subsequent clinical condition of the treated patients, and since the randomized assignment of treatment does not deal with these problems, a predominantly statistical approach to therapeutic trials cannot be expected to resolve the existing scientific controversies about treatment.

Clinical effects of recurrent attacks of acute rheumatic fever: a prospective epidemiologic study of 105 episodes☆

Abstract A large population of children and adolescents with preceding index attacks of rheumatic fever has been followed prospectively at frequent intervals for 11 years in an epidemiologic out-patient research clinic, where special methods were employed to improve the reliability of clinical examination procedures. Despite the antistreptococcal prophylactic regimens assigned to all members of the population, 78 patients have had 105 recurrent attacks of rheumatic fever. During 34 recurrences in 29 patients with no significant murmurs in the index attack of rheumatic fever, 4 patients had seemingly new murmurs, all of which later disappeared leaving no evidence of residual heart disease. Rheumatic recurrences in 6 patients with possible carditis in the index attack also brought no new permanent cardiac damage. By contrast, evidence of new cardiac damage appeared in 48 of the 65 recurrences of rheumatic fever in 43 patients with definite cardiac involvement in the index attack. The new damage was permanent in 11 (37 per cent) of 30 patients whose ‘index’ carditis was manifested only by significant murmurs, and in 5 (38 per cent) of 13 patients with cardiac enlargement or decompensation in the index attack. Although faithful maintenance of antistreptococcal regimens reduced the number of recurrences that would have been expected in such a population before the availability of prophylaxis, the fidelity of previous maintenance of prophylaxis was ‘good’ in 73, and ‘not good’ in 32, of the 105 recurrent attacks. When a recurrence developed, the likelihood of new carditis seemed unrelated to the fidelity of previous prophylaxis and depended mainly on the preceding cardiac state of the host. These results indicate the importance of the host in cardiac effects of recurrences of rheumatic fever. Rheumatic recurrences seldom bring new cardiac damage to hearts previously free of it, but often adversely affect those that have already been injured.

TAXONOMY AND LOGIC IN CLINICAL DATA

As computers become available to help or to hinder progress in clinical medicine, we clinicians must evaluate our own intellectual algorithms-the concepts we use in thinking about what we do in managing the problems of human sickness. We can apply computers merely as a mechanism for automating the status quo, or we can contemplate ways of changing and improving the status quo itself. For this contemplation, we must begin by recognizing what is the status quo, and how did it get that way. The actual events of nature are shown in FIGURE 1. A healthy host, exposed to a cause of disease, becomes a diseased host who undergoes a certain natural course of that disease, becoming an evolved host. During that time, if the diseased host is fortunate or unfortunate enough to come to medical attention, the doctor will apply a name, called diagnosis, to the disease. He will then reason backward to try to discern the cause, or etiology of that disease; he will then reason forward to predict the outcome, or prognosis. From antiquity until several centuries ago, before the advent of modern technology, the components of human ailments were identified according to the concepts shown in FIGURE 2. A host, shown in the large circle on the right, interacted with a cause, shown in the circle on the left, to develop a disease, shown in the overlap of the two circles. Doctors could observe and describe the host in terms of demographic data: such features of the environment as place of birth, climate, and occupation; and such personal attributes as age, race, sex, and education. Doctors could observe and describe the disease in terms of clinical data: the subjective symptoms of the host and the objective signs discerned on physical examination. Since cause was not observed or proved, it was a matter for various kinds of speculation and conjecture.

The use of a computer for data management in large-scale long-term cooperative studies.

Abstract The problems of medical administration and data processing for certain large-scale, long-term cooperative clinical trials in the Veterans Administration have been greatly facilitated with the aid of a digital computer. For certain projects, all incoming data, after initial review by the clinical chief of the project, can be recorded and stored in the computer, which is programmed to check for omissions of data or for egregiously incorrect entries, and to generate notices that call these errors to the attention of the clinicians. Alternatively, in other projects, excerpts of the data can be coded separately on punched cards that are processed in the computer as needed. For each clinic participating in a project, the computer can produce the following material: schedules of patient visits, reminders of procedures to be performed at each visit, monthly accounts of salient current data for each patient, and a variety of special reports. All these reports permit frequent assessment of activities at each clinic, and also provide the project director with an up-to-date account of the status of the entire project. Prompt receipt of such information is particularly important for projects in which the future course of decisions or actions depends on continuous analysis of observed results. For semi-annual meetings of the collaborators, the computer can produce statistical tabulations and other useful summaries. This type of computer processing for carefully collected and catalogued data can be a valuable aid in the administrative procedures needed for effective collaboration of investigators, adherence to the research protocol, and analysis of data in a large-scale multi-center clinical trial.

An empirical comparison of random numbers acquired by computer-generation and from the rand tables.

Abstract The use of Monte Carlo and other statistical procedures that depend on randomization of the observed data has been inhibited by uncertainty about the validity of random numbers generated by a computer. In the empirical test performed here, computer-generated numbers appeared to be just as “random” as those available in the Rand tables that are a conventional standard source of random numbers.