Types of Epidemiological Studies in Peruvian Biomedical Journals

M any decisions made in the daily practice of med- under the headings above(editorials, clinical or icine are based on advances in scientific knowledge, pathological discussions, interviews, book journals being one of the more important means for reviews, translations of foreign articles, etc.). its dissemination. For this reason, journals must provide reliable and truthful information concern- The number of articles for each journal was the ing advances in science. sum of those given in the index. The articles consi- Recently there has been special interest in study- dered to be epidemiological were classified in turn ing the nature, reliability, and of according to the type of epidemiological study " -;", information offered in specialized journals. Thus, the type of statistical analysis used "", and the nature for example, Feinstein "', Fletcher and Fletcher ', of the references cited. and Bailar '-, among others, have designed classifi- The types of epidemiological studies considered cation schemes for biomedical research which make were: it possible to do a critical analysis of studies, to identify the most powerful methods of research, and 1. Descriptive to systematize it. On the other hand, several authors ,4-" have found the frequency with which important - Strictly descriptive: details or characteristics data are omitted related to the design and presenta- of a phenomenon or situation are presented tion of results is very high, even in serious journals without any intention of establishing associ- of worldwide circulation. ations such as case reporting, surveys, stu- In the present study an evaluation has been made dies on , , mortality, of information contained in the regularly published morbidity (209 studies, 71%). Peruvian medical journals of widest circulation, Descriptive of effectiveness: observations on focusing on the type of study published, with the effect of a treatment or procedure in a emphasis on those of epidemiological content, and group of persons without any control group, reviewing the statistical analyses used and the bibli- such as uncontrolled trials (65 studies, 22%). ography cited. The study was initiated and deve- loped almost totally as an undergraduate curricu- 2. Analytical: Studies that attempt to establish an lum in the Comprehensive Health association or causal relation, or to evaluate pro- Course at the Alberto Hurtado School of Medicine, posed hypotheses. Universidad Peruana Cayetano Heredia, Peru. Reviews were undertaken of all the articles - Retrospective: a study of cases and controls printed in the journals Tribuna Médica, Diagnós- in which the primary observation is the tico, Galeno, and Revista de Neuropsiquiatría pub- effect, based on which the imputed factor is lished between 1979 and 1983, plus issues (6) 1-3, (7) traced as the cause (one study). I and 2, (9) 1 and 2, and (1 1) 2 of the journal Acta Cross-sectional: a study that analyzes the Médica Peruana. Each article was classified accord- association between two apparently inde- ing to the type of study, its motivation and the style pendent characteristics or situations ob- of presentation, according to the following hier- served simultaneously, for which the direc- archy: tion of the cause-effect relation is imprecise (9 studies). 1. Epidemiological: original study of manifesta- - Prospective: a study based on the longitudi- tions, therapeutic tests, causality, or nal follow-up of a particular cohort which attributes in groups or populations. attempts to examine the effect of exposure to 2. Technological: study whose principal objective a factor imputed to be causal without any is to demonstrate the validity or merits of a manipulation by the observer (2 studies). technique or procedure (sonography, tomo- Experimental: evaluation of a treatment graphy, etc.). through the comparison of a group exposed 3. Review: article that selects, synthesizes, or to the treatment and a comparable control evaluates information already published, with group selected randomly based on a double- or without examples of clinical cases. blind design (randomized controlled trials) 4. M inor publications: other articles not included (3 studies).

6 - Pre-experimental: a study defined in the standardize the definitions, confirm the type of same terms as the previous type except for epidemiological study, and discuss any mistakes the double-blind design and the random made in the design of the studies. In general there selection of the subjects, such as the non- was no difficulty in reaching agreement. random controlled trials (2 studies). R esults 3. Methodological: Studies of sensitivity and/or specificity (4 studies). Table I shows the overall breakdown of the arti- cles according to the type of study. It is observed The articles were assigned to one of these catego- that most of them are review articles (43%), fol- ries depending on the study design and on the results lowed by minor publications (38%), epidemiologi- obtained, independently of the authors' opinion '' cal studies (15%), and finally, technological articles According to the statistical methods utilized, the (5%). It was found, in addition, that Acta Médica studies were subdivided as: Peruana and Galeno were the journals with the largest percentages of epidemiological studies, and 1. No statistical method or descriptive statistics Diagnóstico and Tribuna Médica, of review arti- only: percentages, means, standard devia- cles. tion, histogram, etc. (257 studies, 83%). Of the epidemiological studies, 93% are descrip- 2. One-sample t test, paired t test, two samples tive, 5% analytical, and 1% methodological. This test, z transformation test (16 sttidies, 5%). same trend was observed for each of the journals 3. Contingency tables, Chi squared test, individually, with the exception of Acta Médica Fisher's ideal index, McNemar's test (7 Peruana (100% descriptive). studies). With regard to the type of statistical analysis 4. Other non-parametric tests of the sign, employed, the great majority of epidemiological Mann-Whitney, Wilcoxon (one study). articles (257) used only descriptive statistics and the 5. Epidemiological statistics: , sensi- 37 remainder ones used a total of 51 statistical tivity, specificity, rates (five studies). methods. The method most used was the t test, 6. Simple linear regression, correlation coeffi- followed by contingency tables. cient (5 studies). Table 1. Breakdown of articles by type of study. 7. Life tables: actuarial, Kaplan-Meier estimate of survival (2 studies). Articles 8. Survival regression: Cox regression and Type of study No. % logistic regression. 9. Analysis of variance, covariance and F test. Epidemiological 294 14.8 10. Multiple regression: polynomial, step-wise. Non-epidemiological 11. Logarithmic transformation to be used in - Technological 91 4.6 another test. - Informative review 852 42.8 - Minor publication 12. Other: any other analysis not mentioned (editorials. interviews, above (calculus of probability, K olmogorov- díscussions) 752 37.8 Smirnov goodness-of-fit test, etc.). Total 1.989 100.0 In articles that used any other statistical method besides the descriptive one, the latter was not consi- Table 2 shows that most of the descriptive epide- dered. Articles which conclude there are significant miological studies do not use statistical analyses and differences between some of the parameters studied that most of the analytical and all the methodologi- but which did not specify the statistical method cal articles do use some type of analysis. used, were classified as descriptive statistics. The classification of articles '- "" was done by two Table 2. Statistical methods by type persons independently; in the case of discrepancies, of epidemiological study. the two reviewers analyzed the article together and No method Some method came to a decision. If not achieved, the article was No. % No. % reviewed by all the authors. Given the special struc- ture of the journal Galeno these procedures were jointly from the start. Descriptive 252 91.9 22 8.1 carried out Analytical 5 29.4 12 70.6 All the articles indicated to be epidemiological Methodological 0 0 4 100.0 were jointly reviewed by four or more authors, to

7 In table 3 it is shown that out of 287 epidemiologi- The differences found between the foreign and cal articles, 175 (61%) do not cite any reference to Peruvian journals in terms of proportions of analyt- national journals, while only 26 articles (9%) lack ical studies are notable; the preponderance of des- references to foreign journals. It is also noted that criptive studies in Peruvian publications places only 4 articles have more than 10 references to them at the most elementary level of scientific national journals. research. While in Peruvian journals only 16% of the Table 3. Characteristics of the references cited epidemiological studies utilize statistical methods in the epidemiological articles. that are not merely descriptive, in foreign journals this proportion rises to 42%, according to Emerson Articles ""', and to 47%, according to Háyden "". This gives Type of reference No. of references No. % an idea of the poverty of the statistical content of the Peruvian publications. As was to be expected, the National None 175 61.0 analytical studies use more statistical methods than 1-4 98 34.1 the descriptive ones; what is notable is that in almost 5-18 10 3.5 Over 10 4 1.4 a third of the analytical articles no statistical analy- sis was used to back up the conclusions. Foreign None 26 9.0 In the analysis of bibliographic references, the 1-10 136 47.4 Over 10 125 43.6 national references are rarely cited, compared with the foreign ones; this perhaps reflects the scarcity of Books None 128 44.6 information widely disseminated in the national 1-10 148 51.6 Over 10 11 3.8 journals, or else, a lack of confidence in them "'."' There is great variability in the definition of Other (theses. work- None 137 47.7 terms, which makes it difficult to compare the shops, congresses) 1-10 147 51.2 Over 10 3 1.1 results of this study with others, but such compari- sons are necessary in order to have a detailed under- Note: 7 articles had no references. standing of the problem. Discussion Finally, although the study design did not allow for studying the quality of the articles, the approp- The aspects reviewed in this study describe the riate use of statistical methods and the minimum profile of information available in Peruvian publi- requirements for the publication of studies, numer- cations. It can be seen that nearly half the articles ous conceptual errors were found in terms of the (42.8%) are reviews of existing information, and design of studies and for interpretation of the data in only a small percentage (14%) have epidemiological the epidemiological articles reviewed. Information content. The possible explanation for this finding on ways to avoid these errors and present a study of could be that the purpose of the publications studied acceptable publication quality is given in the biblio- is either medical education or the dissemination of graphic references '""' scientific knowledge, or that these are the only works available for publication, reflecting perhaps References the lack of means or incentives for research. Of the epidemiological studies, there is a large "' Feinstein, A. Clinical XLIV. A survey percentage of descriptive studies (93%). Fletcher of the research architecture used for publications in and Fletcher '", analyzing the JA MA, Lancet, and N general medical journals. Clin Pharmacol Ther 24: Eng J Med, find that of 155 original articles 22% are 117-125. 1978. descriptive, 8% retrospective, 34% prospective, 10% '2' Fletcher, R. and Fletcher, S. in descriptive of effectiveness, 6% pre-experimental, general medical journals. A 30-year perspective. N and 5% experimental. Moreover, Feinstein '", ana- Eng J Med 301: 180-183, 1979. lyzing 324 articles from N Eng J Med and Lancet '" Bailar, J., Louis, T., Lavori, P. and Palansky, M. found that 40% were descriptive studies. Bailar et al. A classification for biomedical research reports. N '", out of 332 original articles from N Eng J Med, Eng J Med 311: 1482-1487, 1984. found 39% that would correspond to our category "'>DerSimonian, R., Charette, L., McPeek, B. and of descriptive (what they call "cross-sectional"), and Mosteller, F. Reporting on methods in clinical 6%, retrospective. Brown "' found in Am J Dis Child trials. N Eng J Med 306: 1332-1337, 1982. that 34% were prospective studies, 7.2% retrospec- '>'Mosteller, F. Problems of omissions in communi- tive, and 58.5% descriptive. cations. Clin Pharmacol Ther 25: 761-764, 1979.

8 '1' Brown, G. and Baca, G. A classification of origi- "" Research Development Committee. Clinical nal articles. AJDC 1983 and 1984. Am J Dis Child Methods. An annotated bibliography. Ann Intern 140: 641-645, 1986. Med 99: 419-424, 1983. "' Fulginiti, V. Statistics, reviews, and AJDC. Am J Dis Child 140: 627, 1986. '' Brown, G. Statistics and the medical journal. Am (Source: Drs. Carlos Estrada, Jennie Andresen, J Dis Child 139: 226-228, 1985. Adolfo Villar, Justo Montalvo, Juan Limo, Pedro (9 Mac Mahon, B. and Pugh, T. Principiosy méto- Muro, and Diego González. Alberto Hurtado dos de epidemiología. (2nd ed) Mexico, La Prensa School of Medicine, Universidad Peruana Caye- Médica Mexicana, 1975. tano Heredia, Lima, Peru. This article first appeared ""' Emerson, J. and Coidits, G. Use of statistical in Diagnóstico, Vol. 21, No. 6, 1988.) analysis in N Eng J Med. N Eng J Med 309: 709-713, 1983. "" Hayden, G. Biostatistical trends in Pediatrics, implications for the future. Pediatrics 72: 84-87. 1983. Editorial Comment "" Garfield, E. Citation analysis as a tool in journal evaluation. Science 178: 471-479, 1972. The lack of access to scientific health literature "'' Garfield, E. Significant journals of science. adds up to the other difficulties faced by those dedi- Nature 264: 609-615. 1976. cated to activities of health care, training and ('1 Cole, J. and Cle, E. The Ortega hypotesis. Science research, thereby obstructing the general develop- 178: 368-375, 1972. ment of . <"' International Committe of Medical Journals Edi- In the complex process of identification of valid tors. Uniform requirements for manuscripts sub- knowledge generated in the Latin American coun- mitted to biomedical journals. Ann Intern Med 96 tries, very useful-though partial-information (part ll): 758-771, 1982. stems from the revision of national bibliographic ""' McLean, R., Anderson, V., Bayer, H. and McEI- sources. It is necessary to stimulate however, the rath, G. A scientific approach to experimentation critical capability to analize such production, thus for consulting statisticians. J Quality Technol 5: 1-6, making available valid information on health condi- 1973. tions of the population, on health services, and on "7' Feinstein, A. and Horwitz, R. Double standards, the efficiency and efficacy of appropriate technol- scientific methods, and epidemiologic research. N ogy for intervention. Eng J Med 307: 1611-1617, 1982. The use of the for the "J' Altman, D., More, S., Gardner, M. and Pocock, attainment and analysis of such information, as the A. Statistical guidelines for contributors to medical article suggests, is considered an indispensible qual- journals. Br Med J 186: 1489-1493, 1983. ity of a publication.

Epidemiological Activities in the Countries

Analytical Workshop on Functions and Activities Among the participants in the event were profes- of Epidemiological Surveillance in Uruguay sionals from the Division of Epidemiology (Epide- miological Surveillance, Communicable Diseases, On 14-23 December 1988 an Analytical Work- Noncommunicable Diseases, Program for the Con- shop on Epidemiological Surveillance Functions trol of Smoking), other departments of the Ministry and Activities was held in Montevideo under the of Health (Laboratory, Nutrition, Vector Control, sponsorship of the Division of Epidemiology of the Zoonosis, AIDS, Tuberculosis), and the working Ministry of Public Health. group on data processing.

9