[P and beta in statistics].
Explore the source record for details and available documents.
Confidence Intervals↗
Biomedical subjects
J Hilden
Publications and source records attributed to J Hilden.
Observer homogeneity in the histologic diagnosis of Helicobacter pylori. Latent class analysis, kappa coefficient, and repeat frequency.
Four pathologists independently examined 82 antral mucosal biopsy specimens for the presence of Helicobacter pylori and indicated whether their assessments were certain. The pathologists made a positive diagnosis in from 56% to 84% of the specimens (significant heterogeneity, p < 0.01). The frequency of uncertain diagnoses was from 4% to 20% (p < 0.01). Uncertain statements occurred more frequently among negative than among positive diagnoses. For the six pairs of observers the kappa coefficients were between 0.39 and 0.82. By a latent class analysis measures of diagnostic accuracy were calculated comparing the observers' assessments with an estimated consensus diagnosis. The predictive values of a positive diagnosis ranged from 0.70 to 1.00. By calculation of repeat frequencies--that is, the probability that an observer's statement was confirmed by another observer--it became evident that uncertain statements were less frequently (61%) confirmed than were certain ones (85%). It is concluded that observer homogeneity is only moderate with regard to the histologic diagnosis of H. pylori, which should be considered both in daily clinical routine and in scientific studies. Disagreement between observers was associated with negative diagnoses, presumably because the pathologists felt more uncertain in these cases.
Adult↗
[Clinical decision support and informatics].
Explore the source record for details and available documents.
Decision Making, Computer-Assisted↗
[Systems for clinical decision support: a terminological mini-Baedecker].
Explore the source record for details and available documents.
Artificial Intelligence↗
[Observer variation and accuracy in the clinical diagnosis of ascites].
Seventeen observers participated in an observer variation study of the clinical evaluation of ascites. In a blinded design, eight patients with diagnoses of liver disease were examined. Fourteen observers examined the patients twice in order also to estimate the intra-observer variation. The accuracy of the observers' statements was compared with ultrasound findings, by which mean ascites was demonstrated in two patients. Poor correspondance between the observers' gradings and volume estimations, and poor accuracy of the gradings, make qualitative and quantitative ascites estimations useless. The inter-observer agreement was found to be low although the intra-observer agreement was good. The observers' subjective certainty of correctness of their own findings, marked as certain/uncertain, did not reflect the chance of making a correct statement on each particular occasion. The individual patient's general ability of inducing certainty did relate the chance of forming a correct diagnosis. Ultrasonic investigation of the abdomen is recommended in all situations in which demonstration of ascites is essential to diagnosis or therapy.
Ascites↗
Estimation of the spontaneous abortion risk in the presence of induced abortions.
We propose a method of estimating the miscarriage risk in a setting where counts of births, miscarriages and induced abortions are available, and also the gestational week of each induced abortion. Unlike previously proposed methods, ours takes into account the fraction of the miscarriage risk to which each interrupted pregnancy has been exposed. For this purpose, we draw on an extraneous standard foetal survival curve and stipulate that the survival curve of the study population is a scaled version thereof. Three such scaling models are discussed, and it is shown that the choice is largely a matter of computational convenience. Separate attention is given to a competing-risk model of miscarriages and interruptions, and examples are given of reasons why the crucial assumption that these two sources of termination operate independently is unlikely to be met. Finally, it is argued that pregnancy wishes, especially those of habitual aborters, shape the miscarriage rate to the extent that it becomes as much a cultural parameter as a marker of biological hazards.
Abortion, Habitual↗
Ultrasound examination in jaundiced patients. Is computer-assisted preclassification helpful?
In this study we attempted to determine the diagnostic accuracy and reproducibility of ultrasonography (US) for jaundice and to see how US can best be combined with preliminary clinical-biochemical diagnoses to plan the invasive work-up. US proved reproducible in two diagnostic departments (127 agreements in 135 cases). But, since obstruction was underdiagnosed (15 double-false negatives), the predictive value of a negative result was only 0.83. By adding a term which represents the US conclusion, obstruction or not, to the Copenhagen pocket diagnostic chart score (based on the logistic model) we found that an obstructive conclusion increases the odds of obstruction by a factor of 25, and a non-obstructive conclusion decreases the odds by a factor of only 1.9. We conclude that the preliminary diagnosis is frequently sufficiently certain to be unalterable by US. This leaves only 40% of the jaundice cases in which US is necessary to plan invasive work-up. The US workload can even, it appears, be reduced to about 22% without appreciable penalty in terms of unrewarding invasive procedures. Using these strict indications, four US examinations seem to suffice to avoid one such error. Relying on either US or clinical-biochemical data alone is inferior to the combined strategy.
Algorithms↗
The safety of chorionic villus sampling. A synthesis of the literature.
Altogether 10 reports on the safety of chorionic villus sampling, either by the transcervical (TC) or the transabdominal (TA) approach, were reviewed and combined with our own data. After discussion of how unintended fetal loss rates are best estimated, the excess total fetal loss after TC and TA compared with amniocentesis were estimated to be 1.70% (+/- 0.65%) and practically zero (+/- 1.0%), respectively (standard errors in parentheses). The absolute risk of unintended loss after TC is +2.7% (+/- 0.7%) and after TA 1.0% (+/- 1.0%). These estimates are still too uncertain to allow precise weighting of benefits and human costs. A uniform style of reporting studies in this area is proposed.
Abdomen↗
[Confidence intervals instead of p-values].
International scientific journals expect authors of articles to an increasing extent to calculate confidence intervals for their statistical findings. Confidence intervals are more informative than p-values in hypothesis testing as the confidence interval expresses how great the value of an investigated effect may be anticipated to be in the population. Examples of calculation of confidence intervals are presented on the basis of data frequently occurring in medical investigations.
Confidence Intervals↗
Analysing clinical decision analyses.
We present a critical review of aspects of clinical decision analysis which uses an application to screening for familial intracranial aneurysms. The analysis is reported together with methods for assessing decision trees. These methods appear to be powerful checks on the usually rather intuitive way in which decision trees are built. The problem of assessing the uncertainty in the results of a decision analysis is discussed in detail. In practice, sensitivity analysis covers nearly every calculation apart from the standard evaluation of the decision tree. Different forms of sensitivity analysis are distinguished and given appropriate names: influence analysis, threshold analysis, full Bayesian analysis, Bayesian influence analysis, attribute analysis, generalization analysis and scenario analysis. The biostatistical community may well contribute to the much needed methodological improvement in decision analysis and its different forms of sensitivity analysis, especially if prepared to look beyond the standard statistical techniques.
Adult↗
The marriage of clinical trials and clinical decision science.
Clinical decision science is concerned with rational clinical decisions. All branches of medical research contribute here, but controlled clinical trials of the pragmatic variety carry a particular responsibility. Usually, however, they are not conducted and reported so that they can be used directly as input to a decision analysis. We suggest that the forces of the two methodologies should be united, and point out some areas where this 'marriage' will have a non-trivial impact: choice of end points, style of outcome recording, adaptive designs, and style of result presentation. Special attention is given to the decision-analytic setting of research priorities, the role of utility calculus in quantifying the ethical dilemmas that surround clinical trials, and the use of patient attitude towards outcomes of treatment as a covariate in its own right.
Clinical Protocols↗
Which types of perinatal events are predictable? A look at a risk score model.
This study describes the association of a risk factor model for complicated delivery, perinatal morbidity and perinatal mortality with each of various types of delivery complications, types of perinatal morbidity and causes of perinatal mortality. The material comprises a total cohort, 4,066 pregnant women with singletons in a Danish county, and their newborn infants, of whom 494 (12%) had clinical morbidity during the first 5 days of life; 28 (0.7%) died perinatally. A set of 20 risk factors, identifiable before pregnancy, at any time during the pregnancy or at term, was devised by joining existing models for prediction of complicated delivery and of perinatal morbidity and mortality. Metabolic and disproportion-related events were well predicted by the model, inertia-related ones less so, and placental conditions not at all, except for abruption. All types of neonatal morbidity (except sepsis) were well predicted, as were deaths. The strongest predictors of perinatal death were signs of hydramnios (RR = 16.1) and growth retardation (RR = 7.2). The 20 risk factors affected 43% of the population, predicting 57% of the unfavorable perinatal events.
Cohort Studies↗
A computer aid for early diagnostic classification of jaundice (the COMIP program).
We describe the Computer Icterus Program (COMIP), a computer-assisted diagnosis (CAD) program which is designed to facilitate the early choice of a diagnostic strategy in cases of jaundice. To anticipate transfer to other centres, the COMIP program offers facilities to make local adjustments for relative disease incidences and for laboratory data. This is a useful extension for CAD systems.
Algorithms↗
Environmental health hazard handling: statistical and strategic concerns.
Some key issues in industrial and environmental health risk research and the ensuing political decision process are discussed, with particular emphasis on statistical trouble spots: the acceptable risk concept; the welfare of future generations; burden of proof; research priorities; public visibility of risk and benefits; statistical extrapolations; multiple testing. The "tiny thief" phenomenon, discussed in detail, is the paradoxical fact that an almost harmless chemical may steal more of our health than a grossly hazardous chemical because the former is allowed to operate over a disproportionately long time span before being detected, if ever. The difficulty of proving that something is safe is reviewed. Finally, it is urged that politicians be taught the facts of scientific, including statistical, life; likewise, scientists should be made aware of the nature of political decision processes. Otherwise scientists will emigrate from health hazard research into areas of science where equivocal results and political interference are less common.
Biometry↗
The unborn and newborn child. I. Risk factors predicting complicated delivery in a general population of 4,102 women.
Among the 56 risk factors (RF) in pregnant women, used by the Danish National Board of Health, those that can predict complicated delivery (CD) were identified. The significance of parity, maternal age, social class and civil status was also analysed. The material comprises a Danish county cohort of 4,102 deliveries. The 56 original RFs affected 56% of the population. Women (8.8%) with twin pregnancy, fetus in breech, footling and transverse lie, or having an elective cesarean section were analysed separately. The incidence of CD was otherwise 39%. Of all the women, 8.7% had only prepregnancy RFs with a CD rate of 52%; 19.2% had only pregnancy RFs other than special conditions mentioned with a CD rate of 52%; 3.6% had both prepregnancy and pregnancy RFs with a CD rate of 62%. When the special conditions separately analysed were included, 14 RFs of the original 56 were found to predict complicated delivery. These affected 40% of the population. Primiparity was also a RF. The conceptional age of a primipara raised the odds in favor of CD by a factor 1.09 for each year. Social class and civil status were of no significance for CD.
Adolescent↗
The unborn and newborn child. II. Risk factors predicting perinatal morbidity and mortality in 4,138 infants.
In this study, antenatal risk factors (RF) predicting perinatal morbidity and mortality (PMM) were identified among 56 RFs defined by the Danish National Board of Health. The association with parity, age, social class, civil status, complicated delivery was also analysed. The RFs predicting complicated delivery have been described in Part I (1). All events, both prenatal and during the perinatal period, in 4,138 infants borne by 4,102 women in an entire Danish district were analysed. The frequency of perinatal mortality was 0.8% and of perinatal morbidity, 12.7%. The original 56 RFs affected 56% of the population. Fourteen 'new' RFs among the original 56 RFs predicted PMM and affected 27% of the population. The prepregnancy RFs affected 4.5% of all women with singleton pregnancies and their infants had a PMM rate of 21%; the pregnancy RFs affected 18.1%, the PMM rate being 25%; 3.4% had both prepregnancy and pregnancy RFs, their PMM rate was 41%. Twin pregnancies occurred in 0.9%, with a PMM rate of 47%. Apart from these groups, the PMM rate was only 8%. Parity, social class and civil status were of no significance for PMM. Some delivery complications, termed labor RFs, raised the odds of PMM by a factor of 1.92.
Adolescent↗
Risk indicators for low back trouble.
A general population of 928 men and women aged 30, 40, 50 and 60 years participated in a health survey with emphasis on low back trouble (LBT). In all 135 variables were analysed to identify possible indicators for first-time experience and recurrence or persistence of LBT during a one-year follow-up. Stepwise logistic regression analyses were carried out to identify the most informative combinations of indicators for prediction of LBT. For men, a high risk for recurrence or persistence of LBT was associated with frequent LBT in the past, worsening of the LBT since its onset, sciatica and living alone. For women corresponding risk indicators were: recency of the last LBT episode, waking up during night because of LBT, aggravation of LBT when standing, rumbling of "the stomach" and smoking. The strongest risk indicators for first-time experience of LBT were epigastric pain, daily smoking and low isometric endurance of the back muscles. In addition, hospitalisations for whatever cause and a long distance from home to work showed predictive power for first-time LBT among gainfully employed participants. The results indicate that persons with either recurring or first-time LBT had more health problems and probably lived under a higher psycho-social pressure than those without LBT in the follow-up year.
Adult↗
A decision tree for early differentiation between obstructive and non-obstructive jaundice.
We present a method for early differentiation between obstructive and non-obstructive jaundice. On the basis of 14 variables (clinical data and clinical chemical tests, all available within 48 h) a simple decision tree or flow chart has been constructed. The diagnostic yield was as follows: 857 of 982 consecutive jaundiced patients (87%) in a data base and 98 of 108 patients in an independent test sample (91%) were correctly classified. Decision trees for the differentiation between benign or malignant causes within the obstructive group and between acute or chronic causes within the non-obstructive group are also presented. The resulting four-way classification was correct for 77% of the patients in the data base and for 72% of the patients in the test sample. The decision trees are compared with previous methods founded on Bayes' rule and logistic discrimination. The decision trees enable a quick and reliable classification of jaundiced patients, thus providing a valid basis for rational planning of the further diagnostic study.
Cholestasis↗