Richard G. Cornell

National experience with extracorporeal membrane oxygenation for newborn respiratory failure. Data from 715 cases

In a national registry, data were collected on 715 newborn patients with severe respiratory failure supported by extracorporeal membrane oxygenation (ECMO) in 18 neonatal centers. This represents almost all infants treated with ECMO between 1980 and 1987. Eighty-one percent of the patients survived. This result is statistically significantly better than any other treatment which produces less than 78.4% survival. The most common diagnoses were meconium aspiration (310 patients, 91% survived), respiratory distress syndrome (96 patients, 78% survived), diaphragmatic hernia (121 patients, 65% survived), and sepsis (64 patients, 72% survived). Average pre-ECMO characteristics were: age 59 hours; PaO2 42 torr, PaCO2 41 torr, pH 7.40; ventilator settings FiO2 1.0, pressure 45/4 cmH2O, rate 93. Technical complications occurred in 23.1%, and physiologic complications occurred in 65.6%. Results improved with experience. Survival rate for the first ten patients from each center was 73.5% compared to 83.7% for all subsequent patients. Survival rate did not, however, significantly differ after an institutional experience of 20 patients. These observations were made on a large cohort that could not be accumulated at an individual center. These results indicate that ECMO and lung rest is appropriate and successful treatment for newborn respiratory failure unresponsive to other means of management, and that almost all respiratory failure is reversible in near-term neonates.

Stepwise Tests for Multiplicative Components of Interaction

Stepwise procedures based upon likelihood ratio statistics are used to determine the number of multiplicative components in a fixed two-way model. An approximation is presented to obtain percentage points for these likelihood ratio statistics.

Randomized play the winner clinical trials

Tables are presented for the design of randomized play-the winner (RPW) clinical trials to achieve a high probability of selection of the best of two treatments and a balance between the expected number of patients on the inferior treatment and an index of possible patient selection bias. Included is a method of accelerating the divergence of the treatment allocation probabilities as the patient responses accumulate in favor of one treatment over the other. An application of the RPW approach to a clinical trial on the treatment of respiratory distress in newborn infants is discussed. The trial is interpreted in terms of the posterior probability of correct selection given the observed results. A modified design criterion is suggested based upon the selection of a new treatment over a standard control only when the posterior probability of correct seclection is high

A Prospective Study Identifying Risk Factors for Discontinuance of Insulin Pump Therapy

OBJECTIVE To identify characteristics of adult patients at baseline associated with duration of subsequent, continuous, subcutaneous infusion of insulin treatment (pump therapy) of type I diabetes. RESEARCH DESIGN AND METHODS For 6 wk, patients followed a standardized conventional therapy and kept a record of insulin dosages, capillary blood glucose concentrations, and symptomatic hypoglycemia. They were then hospitalized. Additional baseline data were obtained and pump therapy was started. Survival analysis was used to determine the relationship between baseline independent variables or risk factors and duration of pump therapy, which is the dependent variable. RESULTS Of the 68 participants, 33 (49%) terminated pump therapy after an average of 9.9 mo of treatment. Two models (each P < 0.00005) were developed that exhibited a high degree of consistency. Of the 6 variables, 5 were common to both models (HbA1 autonomic neuropathy, mean amplitude of glycemic excursions, frequency of symptoms of hypoglycemia when blood glucose was < 70 mg/dl, and erythema at injection sites). The sixth variable in model 1 (insulin dosage) was replaced in model 2 by a variable, Adult Self-Efficacy for Diabetes, which was obtained on the 33 more recently enrolled patients; this variable related to patient perceptions of self-care behaviors. CONCLUSIONS We found that, at baseline, the presence of a high concentration of HbAx and a low estimation by the patient of their ability to treat the disease portend failure of insulin pump therapy as evidenced by its discontinuation. This effect is accentuated when clinical evidence of autonomic neuropathy is observed. These findings offer guidance in selecting patients with type I diabetes for insulin pump therapy.

Quantifying Gains from Stratification for Optimum and Approximately Optimum Strata Using a Bivariate Normal Model

Abstract In the context of stratified sampling, the estimation variable y and the stratification variable x are assumed to have a bivariate normal distribution with correlation ρ. Optimum stratum boundaries, which vary with ρ, are obtained for Neyman allocation. These boundaries are compared with approximately optimum boundaries determined by the cum. √f rule and by the cum. rule. Gains in precision due to stratification are tabulated, and when the number of strata is small-to-moderate, the cum. √f rule performs somewhat better than the cum. rule. Related results are presented for proportional allocation.

Sequence length for repeated screening tests

Abstract The cost of detecting asymptomatic colonic cancer with a sequence of occult blood tests by the guaiac method is investigated to determine the effect of sequence length. The results are sensitive to the definition of a screen positive and a variety of definitions of a positive screen are investigated. It is found that additional tests, when practical, can be used effectively to reduce the average cost of detection per case and to decrease markedly the probability of a false positive screen without sacrificing high sensitivity even when the prevalence of disease is low. The results hold for a variety of combinations of test and prevalence parameters. This strengthens the conclusion and makes the results more generally applicable.

The evaluation of bioequivalence using nonparametric procedures

Applications of nonparametric methods to the evaluation of bioequiv-alence for two treatments are presented for independent samples and for a crossover design. Included are procedures for testing for equivalence in location, in dispersion, and in general. Also presented are procedures for the calculation of confidence limits. A general strategy for the evaluation of bioequivalence is developed which involves both hypothesis testing and the calculation of confidencelimits for parameters which characterize departures from equivalene.

Interpreting behavioral, biomedical, and psychological relationships in chronic disease from 2 × 2 tables using correlation

Correlations of 0.3 and 0.4 in psychological studies in chronic illness (e.g. diabetes) have been criticized as having little clinical importance. The present article suggests that correlation coefficients be interpreted as binomial effect size displays (BESD) as a means of gleaning the clinical importance of a given correlation coefficient. The BESD is easily understood, computed, and applied to research in chronic disease. An illustration is provided in which a r of 0.2 is shown to correspond to a BESD in which the success rate is improved from 40 to 60%. Using BESD to interpret the meaningfulness of relationships, correlations as low as 0.2 may have clinical importance.

Nonparametric tests of dispersion for the two-period crossover designo

Nonparametric tests of dispersion for the comparison of two treatments in a two-period crossover design are presented and illustrated. The tests developed are asymptotically independent of similar tests for location which use nonparametric procedures.

CONTENT AND SCOPE OF GRADUATE PROGRAMS IN STATISTICS

Publisher Summary This chapter discusses the content and scope of graduate programs in statistics, with emphasis on Mann–Whitney test in statistics. Mann–Whitney test was developed to fulfill a need to analyze ordinal data from applications in biology and from a need to answer substantive questions of importance. Mann–Whitney test is appropriate for the data at hand, namely, differences and sums of possibly highly skewed variables, from crossover experiments. The null hypothesis of equal distribution that it tests describes bioequivalence well. Thus, the range of application of this test continues to grow. An emphasis on applications is important not only in the development of statistical procedures but also in graduate programs in statistics, particularly graduate programs with an applied context such as the health context. There is variability in biological, medical, and social phenomena; this variability often follows stable, reproducible, and predictable patterns that can be modeled mathematically. This can be emphasized in basic courses, both for majors and nonmajors, through sampling experiments and through presentation of data from realistic and important investigations.