Correspondence

CORRESPONDENCE

Objective and method: a delicate relationship

João Luiz Dornelles Bastos^I; Rodrigo Pereira Duquia^II

^IDentist graduated at the Universidade Federal de Santa Catarina - UFSC - Florianópolis (SC). Master's degree students of Epidemiology, Program of Graduate Studies in Epidemiology, Universidade Federal de Pelotas - UFPel - Pelotas (RS), Brazil

^IIDermatologist, Attending Physician at the Residency Service at Santa Casa de Porto Alegre - Porto Alegre (RS). Master's degree students of Epidemiology, Program of Graduate Studies in Epidemiology, Universidade Federal de Pelotas - UFPel - Pelotas (RS), Brazil

^{Mailing Address}

Dear Editor,

Under a strictly methodological point of view, we would like to comment on the article published in the Brazilian Annals of Dermatology (Anais Brasileiros de Dermatologia) by Miot et al.¹

Apparently, the need to simultaneously answer two research problems caused confusion in the authors, since different questions demand different methodologies for their elucidation. The first objective of the work approached the validity of a diagnostic test, so as to answer to the doubt implicitly put in the introduction: 'can some dermatological findings contribute to cardiac risk stratification?'. This question was only partly responded to, since only test sensitivity was reported. However, its specificity of 83.9% was omitted, limiting a more comprehensive evaluation of its validity. Regarding predictive values (PVs), no mention was made to negative PV (41.7%), neither was any consideration that such PV apply only to populations whose disease prevalence approximates that verified in the study (72.4%), according to Bayes's theorem.²

The second objective considered the relation between coronary artery disease (CAD) and earlobe creases ('Is there an association between CAD and certain creases?'). In that sense, it is stated, on the topic 'Casuistry', that the chosen study design was case-control (CC). We agree that such design is adequate, as long as selection of controls in not dependent on interest exposure and reflects the population base that originated the cases.^3,4 Therefore, reading of the work suggests that the design effectively employed was not this one, which puts the presented results of multivariate analyses at stake.⁴ This is so because CC studies begin with case identification, followed by composition of the control group.³ The simple fact that a study contains cases and controls does not characterize it as study of the CC type. Also curious is the ratio of 2.5 cases for each control, something rarely seen in the literature, which probably reveals lack of methodological adequacy for answering their question.

In summary, clear definition of objectives, choice of compatible methods and thorough description of them are needed for a higher quality in the execution and publication of results of scientific research.

Making ourselves available for further debate, we sign.

REFERENCES

1. Miot HA, Medeiros LM, Siqueira CRS, Cardoso LC, Gumieiro JH, Pandini Filho MA, et al. Associação entre doença arterial coronariana e as pregas lobular diagonal e anterotragal em homens. An Bras Dermatol. 2006;81:29-33.

2. Kirkwood BR, Sterne JAC. Essential medical statistics. 2nd. Oxford: Blackwell Science; 2003. p. 429-46.

3. Kelsey JL, Whittemore AS, Evans AS, Thompson WD. Methods in observational epidemiology. New York: Oxford University Press; 1996. p. 188-213.

4. Grimes DA, Schulz KF. Compared to what? Finding controls for case-control studies. Lancet. 2005;365:1429-33.

Author's answer to the correspondence 'Objective and method: a delicate relationship'

Hélio Amante Miot

PhD, Assistant Professor at the Department of Dermatology at FMB - Unesp - Botucatu (SP), Brazil

^{Mailing Address}

Dear Editor,

The authors have proposed to study the prevalence of certain dermatological findings in patients submitted to cineangiocoronariography and not to the efficiency of a diagnostic method.

Lack of statistical association, in this population, between coronary disease and androgenetic alopecia, as all as thoracic pilification, lead the discussion of results to an analysis of earlobe creases.

The work under discussion is the result of a first attempt to characterize, in the Brazilian population, these dermatological findings using cineangiocoronariography, once a previous nationwide study compared the presence of diagonal earlobe crease between coronary disease patients and outpatients with non-cardiac complaints, which has not estimated absence of coronary disease in controls.¹

Despite the accuracy of the method, the invasive nature of cineangiocoronariography ethically hinders studies in populational bases, which leaves as patients' origin the indication of the exam by cardiac complaints (hospital source). This feature, inherent to the studied population (exams performed in the period of study), would yield a ratio of one patient with normal coronary arteries to each 55 with affected arteries.

A cross-sectional study of prevalence, with no selection of cases or controls, would yield yet a larger disproportion than the one presented between affected and non-affected. For this reason, the authors initially dedicated to the recovery of the largest possible number of 'normal' patients, among the examined.

The authors have not identified the implication of the fact that control selection occurred prior to classic case selection, nor have they observed the methodological constraint, given the sample limitations.

Another issue hovering over literature on the prevalence of earlobe crease in coronary disease patients is the increase in its prevalence with age, parallel to the increase in artherosclerotic risk, which leads to questioning of the role of age as confusion variable for the association between coronary disease and such creases. Answer to that question would demand a stratified populational analysis, or an adjustment by logistic regression.

The option of pairing cases with controls was considered, but sample contingencies prevented that all independent variables were contemplated, thus, the conditional form of multiple logistic regression was chosen, for it is, by definition, more rigorous.²

Based on the assumption that a case-control type of study was more appropriate for the detection of an association between risk factors and diseases, we used this design to analyze the association between variables associated to risk, as well as their magnitude, obeying the formal demands that cases and control come from the same population, which, in this case, was composed by adult patients submitted to cineangiocoronariography.² Other studies on prevalence have employed a similar methodology for either selection or analysis of patients, even though the authors agree that estimative of performance parameters is not fit for this type of study design.^3,4

The possibility of causality analysis between variables was never considered to be among the authors' hypotheses, as classically assume studies of the case-control type. Neither were Hill's criteria evoked, as it is theorized that dermatological findings (earlobe creases) occur parallel to coronary artherosclerosis.

Another important element, as based on previous studies, estimated sample size was more modest than the one actually employed, and it could, by the way, maintain a 1:1 ratio between cases and controls. The choice of a larger number of cases was made in an attempt to increase flexibility of the logistic model and strengthen statistical analysis in face of a low availability of controls in regards to cases.

It is very important to highlight that generalization of results is limited only to the studied population, in this case, hospitalized patients with indication of cineangiocoronariography; nevertheless, internal validity of results must always be prioritized in any study, before extrapolations.

Classically, studies which employ hospitalized controls present difficulties inherent to the selected population, in what regards identification of risk factors for diseases, mainly owing to high exposure levels of controls to risk factors related to other diseases.

Interference of this bias related to studies with hospital base could be observed in this work for the fact that risks such as diabetes, high arterial blood pressure, smoking and family history were not identified. Even body mass index was higher in controls. These known risk factors could probably be detected in studies using a non-hospitalized base.

In other works that involved cineangiocoronariography, employing different hospital populations, results were similar to the ones found in our own, corroborating its validity. Moreover, works that have not used a hospital base, being, rather, based on data from autopsies, confirmed the findings.⁵

As a consequence of this reflection, a significant identification of earlobe creases in this study suggests an even more consistent association when compared to general population, which, however, still needs proving.

Complete lack of orthodoxy in this study demands attention for the interpretation of its data, but does not invalidate its findings, and stimulates its proving by assays conducted by Brazilian research groups.

Still, the authors give incentive for dermatologists to carry out controlled epidemiological studies, preferably adopting the classical models established by modern epidemiological science.

REFERENCES

1. Tranchesi Junior B, Barbosa V, de Albuquerque CP, Caramelli B, Gebara O, dos Santos Filho RD, et al. Diagonal earlobe crease as a marker of the presence and extent of coronary atherosclerosis. Am J Cardiol. 1992;70:1417-20.

2. Hennekens CH, Buring JE. Epidemiology in medicine. Little, Boston: Brown and Co.;1987. p. 132-52.

3. Weiss NS. Application of the case-control method in the evaluation of screening. Epidemiol Rev. 1994;16:102-8.

4. Grandhe NP, Bhansali A, Dogra S, Kumar B. Acanthosis nigricans: relation with type 2 diabetes mellitus, anthropometric variables, and body mass in Indians. Postgrad Med J. 2005;81:541-4.

5. Edston E. The earlobe crease, coronary artery disease, and sudden cardiac death: an autopsy study of 520 individuals. Am J Forensic Med Pathol. 2006;27:129-33.

Publication Dates

• Publication in this collection
  21 June 2007
• Date of issue
  Aug 2006