ORIGINAL ARTICLE

 

Research methodology topics: Cross-sectional studies

 

 

Juliana Zangirolami-Raimundo; Jorge de Oliveira Echeimberg; Claudio Leone

Laboratório de Delineamento de Estudos e Escrita Científica. Centro Universitário Faculdade de Medicina do ABC, Santo André, SP, Brazil

Correspondence

 

 

---

ABSTRACT

In health the most frequent researches are done in the form of observational studies. In this type of scientific research the researchers do not interfere with the phenomena under study, only observe in a systematic and standardized manner, collecting and recording information, data or materials that spontaneously occur at a particular time of the health-disease process, or along its natural evolution, and then proceed with its description and/or analysis. In observational studies normally four types of study design are used: case series studies, cross-section studies, case-control studies and cohort studies. Thus, cross-sectional studies are very useful in descriptive studies when used in studies that are proposed to be analytical, the results must be interpreted by researchers with good experience in that specific field of knowledge, using a lot of caution and common sense.

Keywords: observational study, cross-sectional studies, study design, prevalence.

---

 

 

INTRODUCTION

In health, whether in public health or medicine, the most frequent surveys are done in the form of observational studies. In this type of scientific investigation researchers do not interfere in the phenomena under study, they only observe them in a systematic and standardized way, collecting and registering information, data or materials (such as blood, biopsies and other examples) that occur spontaneously at a given moment in the health-disease process, or during its natural evolution, and then proceed to its description and/or analysis¹.

The aim of cross-sectional studies is to obtain reliable data that make possible to generate, robust conclusions, and create new hypotheses that can be investigated with new research.

Analytical studies seek to establish relationships and associations between two or more phenomena (called variables in the analysis process), and descriptive studies are only about the detailed and organized description of one or more phenomena, the quality of data required for study in addition to the systematization and standardization of the collection methods, also the strategy adopted to obtain them, which is called the design or, more correctly, the study design.

In observational studies normally four types of study design are used²:

- Series of cases,

- Cross-Sectional,

- Case-control and,

- Cohort Studies.

These designs, which has been used in early research, and also perhaps the most frequently used is the cross-sectional study.

The main characteristic of cross-sectional studies is that the observation of variables, whether they are cases, individuals, or other types of data, is performed in a single moment (the same), when the researcher records a "photograph" of the facts (variables) of interest and not the "movie" of its evolution^1,2. Also called transversal, sectional, or prevalence, the cross-sectional study has the advantages of allowing the direct observation by the researcher of the phenomena to be investigated, of performing the information collection in a short time (in public health is very frequent the use of collective effort), without the need for follow-up of the participants, and to produce faster results, therefore, at a lower cost than the other designs^1,3.

These characteristics make cross-sectional studies particularly useful for studying the prevalence^a of a particular phenomenon, whether it is assumed to be the cause or the consequence, or both, in a defined population. These studies, even if purely observational and descriptive, are very useful in the field of Public Health. Of course this type of design is appropriate for problems with prolonged or chronic evolution and is not generally suitable for the study of acute situations, when the interest is the incidence^b of new events.

Beyond the pure description of phenomena, cross-sectional design is also useful in studies that investigate causal and effect relationships, which seek, at least preliminarily, to analyze the relationships between risk factors, determinants and what are supposed to be their consequences or effects (outcomes), such as diseases, sequelae and damages or even advantages (protection) of any kind.

Examples of cross-sectional studies, probably one of the most popular, are population censuses (demographic), systematically carried out by many countries in order to identify characteristics of their populations at a given time, analyze their evolution over time, and to establish some relationships between these features that deserve to be analyzed^3,4.

A census involves a lot of time and resources because it covers the collection of data from the entire population (the universe) that is to be evaluated and, therefore, makes its routine use practically unfeasible in most scientific research.

As a consequence, to reduce the cost and time of conducting the survey it is often necessary to use samples (from the universe) that, based on statistical analyzes, produce results and estimates capable of producing generalizable conclusions, even with some limitations⁴.

For this to be possible it is necessary that the sample used in the research be as representative as possible the study universe, be accurate^c, and that its size (n) is sufficient to guarantee results with the necessary precision^d. Accuracy and precision are not equivalent, because depending on the sample type and its size (n), it can be accurate but very precise or imprecise, but little accurate. The ideal condition is to obtain from the sample an accurate result that, at the same time, is also precision³.

Unfortunately, in many research situations it is impossible to obtain a representative sample, so when this is the case, one uses the best sample that can be obtained, always seeking to have a minimum of representativeness. This type of sample is named the convenience sample.

In those circumstances, despite the fact that the use of a convenience sample does not completely prevents the conduct of a study, it should be in mind that this situation imposes limitations on the interpretation of results, particularly with regard to the generalization of its findings, because it is practically impossible to assess its accuracy.

The study populations (or universes) can either be a population of a city, a state or a country as a whole, as the population of certain subgroups, for example the female or child population, or the population with a certain health problem, such as the population of diabetics or hypertensives, or with exposure to risks such as smokers, sedentary and so on. Researchers must define, from their research hypothesis, which population is to be studied, that is, the one from which a sample will be selected. Whenever it is not possible to obtain a representative sample, the researcher will be analyzing data of what is called the hypothetical population^e.

When it is desired to go beyond purely descriptive aspects, seeking to identify relationships that may exist between study variables, such as the relationship between risk factors and their possible consequences, such as diseases, damages, sequelae, etc., the cross-sectional is considered analytical, establishing comparisons between subjects of the sample exposed to a certain risk factor and those not exposed, whether or not they have a particular disease. Conversely, the study also allows comparisons between patients and non-patients, who were exposed to an alleged risk factor⁵.

Thus, for this analysis it is possible to divide the subjects of the sample, according to the risk factor and outcome (disease), into four distinct groups:

- those who have the risk factor and have the outcome;

- those who have the risk factor and do not have the outcome;

- those who do not have the risk factor and have the outcome and;

- those who do not have the risk factor and do not have the outcome.

In addition, it is possible to divide them into four subgroups based on the disease:

- those who have the outcome and have the risk factor;

- those who have the outcome and do not have the risk factor;

- those who do not have the outcome and have the risk factor;

- those that do not have the outcome and do not have the risk factor.

The distribution of the research subjects is used to make this evaluation, according to the absolute and relative frequency with which they are classified in relation to the two characteristics (risk factor and outcome), in tables named of contingency (Table 1), or association, which are always structured this way:

 

 

In these tables, the data of the boxes (a, b, c and d) are compared with those expected from a distribution that is totally determined by chance. The less divergent the values obtained in the research under analysis from those expected by chance, the lower the probability that there is an association between risk and disease. On the contrary, the greater the divergence of the observed than the expected by chance, the greater the probability that there is an association between the risk factor and the outcome.

The Chi-square methods and the Fisher's exact test are used in this association analysis because they are based on the diference between the expected values and those observed in the research to define their statistical significance⁶.

In this sense it is important to remember that a statistically significant association simply means that there is a probability of being associated, not absolute certainty. In addition, the presence of this probable association does not necessarily imply a relation of determination, that is, of the cause/effect type. An example of this may be the finding, based on a sample of adults, of a statistically significant association between overweight and hypertension. Can obesity lead to hypertension? It may seem true, however there may be another cause, or risk factor, that may be the cause of both: anxiety. An anxious individual may manifest hypertension and become obese, making these two outcomes appear to be one cause and the other the outcome. If the sample is not raised also considering the presence or not of anxiety in the patients , at leastthere is a risck of concluding that hipertension is a case of obesity.

In addition to statistical associations, this type of analysis also allows us to calculate the Prevalence Ratio (PR), defining whether and in which group it is higher. Assuming that it was the same, the result of the division (ratio) of the prevalence of the risk group by that of the risk-free group would be "1". Any significant result other than "1" may indicate that there is an association. If the association is of risk the quotient will be greater than "1", or if it is a protection factor it will be less than "1". Because it is an analysis from a sample, it is difficult to obtain an exact value of "1", which is why the confidence interval (CI) of the obtained value is also estimated. When the value "1" does not fall within the limits of the CI, it means that there is a difference (statistic) between the two groups and, when the value "1" is part of the possible results described by the CI, it is admitted that there is no significant difference between the two groups, exposed and not exposed to the possible risk factor^5,6.

In this model it is also possible to analyze from the outcome, calculating the ratio (quotient) between the proportions of individuals who have exposure to the risk factor in the group with the outcome, the disease, for example, and the group that does not have the outcome. The calculated result is called Odds Ratio (OR), an estimator that is used as an approximation to Relative Risk (RR). When the proportions of the two groups are equal, the result of the OR will be "1", indicating that there is no association between the disease and the exposure to the factor that was supposed to be at risk. The interpretation of other OR values follows exactly the same logic described for PR, now based on the calculated CI for OR.

The cross-sectional design can also be used for multi-variable analyzes, such as binary logistic regression, in order to simultaneously calculate the ORs of several risk factors and their statistical significance, ranking them by order of influence on the outcome under analysis. Moreover, in this analysis it is also possible to calculate the significance and explanatory capacity of the model generated by the set of factors that evidenced statistically significant ORs in the binary logistic regression analysis.

When the phenomena to be studied (risk and outcome) are quantitative (numerical) variables, it is also possible to analyze their relations by means of comparisons of means or medians using parametric statistical tests (Student's T, Fisher's Exact) or non-parametric (Mann- Whitney). In these conditions it is also possible to calculate correlation coefficients (Pearson or Spearman) and, possibly, linear regression models and even to evaluate sensitivity, specificity and predictive values (+ or -) using ROC Curves (Receiver Operating Characteristics Curves).

Despite the potential advantages: faster realization, lower cost, lower losses and the possibility of direct observation of phenomena to be analyzed (avoiding the bias due to memory leaks or inadequate recording or absent stunted information), and enable a wide variety of alternative methods that can be used to statistically analyze data, cross- sectional design (such as the Del Ciampo et al⁷, Lucena et al⁸ study) presents some important drawbacks that make it difficult to interpret association results.

The disadvantages are fundamentally due to the fact of collecting information about risk factors and outcomes are collected in a single moment (the same), which makes it difficult to analyze associations to assess possible cause / effect relationships.

The coexistence of the possibilities of association between risk and outcome, previously described, at the same time (that of the research) nullify the temporal relation that must exist between cause and effect, since it is a sine qua non condition that the cause always precedes the outcome, for this to occur. This can result in the phenomenon of reverse causality. Supposing a sample of the population to observe an association between obesity and joint pain in the lower limbs, it would be valid to suppose that the overload of the weight on the joints would be causing lesions, even if small ones, that would cause the pains. However, it would not be wrong to hypothesize also that chronic joint problems with major pain could condition a reduction in daily activities and a sedentary lifestyle that could lead to obesity.

Another disadvantage also arising from the design is the almost impossibility of applying it when it is desired to approach situations in which the outcome or the risk factor or both are rare in the population, since it would require a very large sample to obtain the number of carriers of the disease required to perform the association analyzes, which at least would increase the time and also the costs for its realization⁵.

As the whole, despite the difficulties, remain as final considerations that the cross-sectional, design can be very useful to assess the frequency of risk behaviors and / or exposure to risks, necessary for the development of public health policies, can also serve as a basis for calculating sample size in the planning of future analytical research with more robust designs to assess cause and effect hypotheses, such as cohort or case-control¹.

 

FINAL CONSIDERATION

Cross-sectional studies have their great use in descriptive studies, while used in studies that are proposed to be analytical, the results must be interpreted by researchers with good experience in that specific field of knowledge, using a lot of caution and common sense.

 

REFERENCES

1.Kramer M.S. Clinical Epidemiology and Biostatistics. Berlin: Springer-Verlag, 1988.         [ Links ]

2. Katz M.H. Study Design and Statistical Analysis. New York: Cambridge University Press, 2006.         [ Links ]

3.Hennekens C.H. and Buring J.E. Epidemiology in Medicine. Boston: Little, Brown and Company, 1987.         [ Links ]

4.Porta M. A Dictionary of Epidemiology 5th ed. New York: Oxford University Press, 2008.         [ Links ]

5.Friis R.H. and Sellers T.A. Epidemiology for Public Health Practice 4th ed. Sudbury Massachussetts, 2009.         [ Links ]

6.Kirkwood B.R. and Sterne A.C. Essential Medical Statistics 2nd ed. Malden, Massachusetts, 2006.         [ Links ]

7.Del Ciampo LA, Louro AL, Del Ciampo IRL, Ferraz IS. Characteristics of sleep habitsamong adolescents living in the city of Ribeirão Preto (SP). J Hum Growth Dev. 2017; 27(3): 307-314. DOI: http://dx.doi.org/10.7322/jhgd.107097        [ Links ]

8.Lucena KDT, Deninger LSC, Coelho HFC, Monteiro ACC, Vianna RPT, Nascimento JA. Analysis of the cycle of domestic violence against women. J Hum Growth Dev. 2016; 26(1): 139-146. http://dx.doi.org/10.7322/jhgd.119238        [ Links ]

 

 

Correspondence:
escritacientifica@fmabc.br

Manuscript received: June 2018
Manuscript accepted: October 2018
Version of record online: November 2018

 

 

a Prevalence: is the proportion of a given population that at the moment of the study presents a disease, risk factor and/or other type of problem.
b Incidence: corresponds to the proportion of new cases (disease, risk or any problem) that occur in the study population over a defined period of time and not just in a single moment.
c Accurate: is the result obtained from a sample, provides a very close value to the one to be estimated in a given population. The accuracy does not depend on a greater or lesser precision of the estimated value, but on the sample representativeness and reliability of the data collected.
e Hypothetical population: the one to which the study's conclusions apply, in other words, if there are populations similar to those in the sample, the conclusions will be valid for them.
d Precision correspond to the variation that a result would have if it were calculated from several samples (from the same population) selected in the same way.The confidence interval (CI) is the one way to represent its precision. The smaller the difference between the extremes of the CI, the greater the precision of the result. Precision does not have to do with the representativeness of the sample, but it depends very much on the size of the sample, that is, on the number of subjects that compose it.