On Some Analyses

On Some Analyses

extra ventilatory efforts or some clinical situation in which the metabolic production of CO2 was decreasing over time. Randolph P. Cole, MD Holy Name...

57KB Sizes 0 Downloads 45 Views

extra ventilatory efforts or some clinical situation in which the metabolic production of CO2 was decreasing over time. Randolph P. Cole, MD Holy Name Hospital Teaneck, NJ Reproduction of this article is prohibited without written permission from the American College of Chest Physicians (www.chestjournal. org/misc/reprints.shtml). Correspondence to: Randolph P. Cole, MD, Holy Name Hospital, 718 Teaneck Rd, Teaneck, NJ 07666; e-mail: [email protected]

Table 2—Some Results of Mean Comparisons From Table 2*2 Measurements

Treprostinil vs Placebo

Heart rate change PAPm change Mean right atrial pressure change Mean pulmonary capillary wedge pressure change

S S S S

*PAPm ⫽ mean pulmonary arterial pressure. See Table 1 for abbreviation not used in the text.

References 1 Demers B. The oximeter: boon or bane. Chest 2004; 126: 1399 –1401 2 Sullivan SF, Patterson RW, Papper EM. Arterial CO2 tension adjustment rates following hyperventilation. J Appl Physiol 1966; 21:247–250 3 Vance JW, Fowler WS. Adjustment of stores of carbon dioxide during voluntary hyperventilation. Dis Chest 1960; 37:304 –313 4 Brandi G, Clode M. CO2 washout during hyperventilation in man. Respir Physiol 1969; 7:163–172

On Some Analyses To the Editor: There are methodological errors being made in statistical analyses, resulting in flawed results. Examples are in the studies by Hiasa et al1 and Oudiz et al.2 The problem is the authors’ use of two-sample t tests, analysis of variance, or analysis of covariance to compare means, which assumes the normality and equality of unknown variances in the groups considered. The Central Limit Theorem justifies normality for mean inferences, but unknown variances need not be equal, making these methods not generally applicable to comparing means. This problem is not removed by futilely3 testing for the equality of variances. Avoiding normality and nuisance variances with rank tests such as the Wilcoxon test2 means that, if significant, they do not specifically say anything about the mean, median, mode, or any specific moment of the distributions, being a comparison of distributions. Moreover, these rank tests are biased4 to one side in a two-sided test. This is irrespective of deaths, the reason given.2 Table 2,2 judging by the labels used, the confidence intervals for mean differences, and the ensuing discussion, clearly shows an interest in means. The confidence intervals are meaningless because the statistic used was not given. The problem of comparing the means of normal populations exactly with unknown variances is the Behrens-Fisher problem,

Table 1—Some Results of Mean Comparisons From Table 2*1 Procedure

Group A vs Group C

Group B vs Group C

I-MIBG early H/M ECG RV5 ⫹ SV1 Echocardiography LVDd

S S

S S S

123

*S ⫽ significant difference between means at 0.02 (one significant figure) significance level; RV5 ⫽ voltage of R wave in lead V5; SV1 ⫽ voltage of S wave in lead V1; LVDd ⫽ left ventricular end-systolic dimension; MIBG ⫽ metaiodobenzylguadine; H/M ⫽ heart/mediastinum ratio. 1888

which was solved by Tsakok5 in its generalized form. The Tsakok solution is more effective in detecting significant mean differences, even with unknown equal variances. Its exposition6 is available elsewhere. A statistical software package (GSP; London, UK) implements the Tsakok technique. Some results from Table 11 and Table 22 are given. These appear to have been overlooked. After taking care to obtain the data, they deserve correct analysis. The article by Tsakok7 on exact, unconditional, uniformly most powerful unbiased tests extends the Tsakok technique to the nonparametric problem of comparing distributions, superseding rank tests or the Fisher exact test (which is neither exact nor unconditional). An extension to dependent samples8 is indicated. The Tsakok articles are reprinted9 with further results. Arthur D. Tsakok, MSc Mathematical Centre London, UK Reproduction of this article is prohibited without written permission from the American College of Chest Physicians (www.chestjournal. org/misc/reprints.shtml). Correspondence to: A.D. Tsakok, Mathematical Centre, 46 Leighton Gdn, London NW10 3PT, UK; e-mail: [email protected]

References 1 Hiasa G, Hamada M, Saeki A, et al. Cardiac sympathetic nerve activity can detect congestive heart failure sensitively in patients with hypertrophic cardiomyopathy. Chest 2004; 126:679 – 686 2 Oudiz RJ, Schilz RJ, Barst RJ, et al. Treprostinil, a prostacyclin analogue in pulmonary arterial hypertension associated with connective tissue disease. Chest 2004; 126:420 – 427 3 Kendall MG, Stuart A. The advanced theory of statistics (vol 2). London, UK: Charles Griffin and Co, 1973; 484 4 Lehmann, E.L. Testing statistical hypotheses. New York, NY: John Wiley and Sons, 1959; 187 5 Tsakok AD. A solution to the generalized Behrens-Fisher problem. Metron 1978; 36:79 6 Tsakok AD. Comment on visual acuity. Ophthalmic Epidemiol 2002; 9:347 7 Tsakok AD. A test of fit satisfying some optimality criteria non-asymptotically. Metron 1978; 36:105 8 Tsakok AD. A generalization of the Borel-Cantelli lemma. Metron 1995; 53:25 9 Tsakok AD. Statistics and the unified field. London, UK: Mathematical Centre, 1987

Function of the Gu¨nther Tulip Vena Caval Filter To the Editor: As an interventional radiologist with a strong interest in vena caval filters, I read the evidence-based guidelines of the Seventh Communications to the Editor