<p>Designing a study forms a vital aspect of conducting a study. A poorly designed study cannot yield results and defects cannot be corrected once study is initiated. Knowledge of a study design is needed not only for conducting study but also for interpreting the study. Study designs are broadly classified into two groups: descriptive and analytical studies. A detailed classification is provided in <xref ref-type="table" rid="tab001">Table 1</xref>. <table-wrap id="tab001"><label>Table 1</label><caption><p>Types of study designs</p></caption><table><tbody><tr><td><list list-type="order"><list-item><p>Descriptive studies <list list-type="alpha-lower"><list-item><p>Case reports</p></list-item><list-item><p>Case series</p></list-item><list-item><p>Cross sectional surveys</p></list-item><list-item><p>Correlation studies</p></list-item></list></p></list-item><list-item><p>Analytical studies <list list-type="alpha-lower"><list-item><p>Observational studies <list list-type="roman-lower"><list-item><p>Case control studies</p></list-item><list-item><p>Cohort studies</p></list-item></list></p></list-item><list-item><p>Interventional studies (Clinical trials) <list list-type="roman-lower"><list-item><p>With controls – randomized or non randomized</p></list-item><list-item><p>Without controls</p></list-item></list></p></list-item></list></p></list-item></list></td></tr></tbody></table></table-wrap><table-wrap id="tab002"><label>Table 2</label><caption><p>Cross sectional studies</p></caption><table><tbody><tr><th>Advantages</th></tr><tr><td><list list-type="order"><list-item><p>Fast and inexpensive</p></list-item><list-item><p>Can study multiple outcomes simultaneously</p></list-item><list-item><p>Yields prevalence rate</p></list-item><list-item><p>No waiting period and no follow up loss</p></list-item><list-item><p>Useful for examining network of causal links.</p></list-item></list></td></tr><tr><th>Disadvantages</th></tr><tr><td><list list-type="order"><list-item><p>Temporal relationship between exposure and outcome cannot be established</p></list-item><list-item><p>Not useful for rare diseases</p></list-item><list-item><p>Incidence is not established</p></list-item></list></td></tr></tbody></table></table-wrap><table-wrap id="tab003"><label>Table 3</label><caption><p>Case control study</p></caption><table><tbody><tr><th>Advantages</th></tr><tr><td><list list-type="order"><list-item><p>High yield of information from relatively few subjects</p></list-item><list-item><p>Useful for study of rare diseases</p></list-item><list-item><p>Multiple predictors of a disease can be studied and hence useful in generating hypothesis</p></list-item><list-item><p>Short duration of study, easy and relatively inexpensive</p></list-item><list-item><p>Yields odds ratio</p></list-item></list></td></tr><tr><th>Disadvantages</th></tr><tr><td><list list-type="order"><list-item><p>Doesn't estimate incidence or prevalence</p></list-item><list-item><p>Only one outcome is studied</p></list-item><list-item><p>Potential for bias is very high and can weaken study</p></list-item></list></td></tr></tbody></table></table-wrap><table-wrap id="tab004"><label>Table 4</label><caption><p>Cohort study</p></caption><table><tbody><tr><th>Advantages</th></tr><tr><td><list list-type="order"><list-item><p>Incidence is obtained and relative risk is calculated</p></list-item><list-item><p>Temporal sequence and causative association can be established</p></list-item><list-item><p>Measures exposure more accurately and bias is minimal</p></list-item><list-item><p>Can study multiple outcomes</p></list-item></list></td></tr><tr><th>Disadvantages</th></tr><tr><td><list list-type="order"><list-item><p>More expensive</p></list-item><list-item><p>Longer duration of study</p></list-item><list-item><p>Design is not useful for rare diseases</p></list-item></list></td></tr></tbody></table></table-wrap><table-wrap id="tab005"><label>Table 5</label><caption><p>Randomized control trial</p></caption><table><tbody><tr><th>Advantages</th></tr><tr><td><list list-type="order"><list-item><p>Strong evidence of cause effect</p></list-item><list-item><p>protection against bias</p></list-item></list></td></tr><tr><th>Disadvantages</th></tr><tr><td><list list-type="order"><list-item><p>Costly in time and money</p></list-item><list-item><p>Ethical barriers</p></list-item></list></td></tr></tbody></table></table-wrap></p></sec><sec id="sec002"><title>Descriptive studies

They are the simplest study designs. They do not have hypothesis and often no comparison of groups. However the simple design, low cost and minimal ethical concerns make them a useful tool as initial step and they pave the way for generating hypothesis to be tested by other study designs.

Case report and case series

It is the most basic type of descriptive study. A detailed description of a single patient or a group of patients with similar pattern of manifestation is provided. They do not involve any study hypothesis and often do not have any controls. Though they do not describe a cause–effect relationship, they may form a hypothesis upon which further studies can be planned.

Many examples of case series can be found in Indian literature. A series of 3 babies with hypernatremia reported in Indian pediatrics is one such example (1). Authors have given a detailed description of these 3 cases. Authors concluded that evaluation of neonates at risk at two weeks of age may help in preventing the devastating short and long-term consequences of hypernatremia. The risk factors contributing to this disorder can now be evaluated by case–control designs.

Cross sectional studies

They are also called surveys. Here, data is collected from a sample population at a point time with no follow up, hence sometimes called prevalence study. The data collected includes both exposure and outcome at same time. It differs from observational studies in that it is cross sectional in time frame while both case control study and cohort studies are longitudinal. Cross sectional studies are suitable designs for describing distribution of multiple variables. As both exposure and outcomes are measured simultaneously, temporal relationship between them cannot be evaluated. Nevertheless, it can establish association which can raise hypothesis to be tested by other suitable study design. An example of a multicentric cross sectional study is by Lokeshwar et al (2). They studied prevalence of varicella zoster antibodies in Indian population. They noted that a significant proportion of adult population is still susceptible. However, the situation may change over years with more use of vaccine or if vaccine use remains low, the susceptible population may increase. Thus cross sectional studies give a situation at a point of time and not the trend.

Serial surveys are cross sectional surveys on the same population observed at several points in time and can give inference about changing patterns over time.

Correlation studies

Here, data from an entire population is used to compare disease frequencies between different groups during the same period of time or in the same population at different points of time. They refer to populations and not individuals, hence exposure to individual factor may not be available and only difference in outcome would be available. Thus, these data also generates hypothesis but do not test them. A study by Avery et al (3) wherein they compared neonates of 8 different NICU for chronic lung disease (CLD) is an example. They noted incidence of CLD to be lowest in NICU of Columbia. A high use of CPAP in this NICU was noted. Birth weight, race and sex difference didn’t contribute to this difference.

Analytical studies

These are the study designs used to test research hypothesis. The basic characteristics of an analytical study are:

Presence of study hypothesis or objectives

Defined methods of selection of study participants

Valid, reliable and reproducible observational methods

Appropriate handling of incomplete observations

Various analytical study designs can be grouped into two broad categories namely observational and interventional studies. This grouping is primarily based on the role of the investigator. In observational studies the investigator plays a passive role and simply observes the natural course of events between exposure and outcome. In interventional studies, also called clinical trials, the investigator controls the exposure and follows up for outcome.

Observational studies

The two basic designs in this category include case control and cohort studies.

Case control study

In case control study, the investigator selects the group with outcome of interest i.e. cases and selects appropriate control group without outcome of interest. He then looks backward and collects data about exposure to risk factor. The exposure measurement is typically retrospective. The value of case control study lies in selection of control group without any bias i.e. a group which matches cases in all known confounding variables. Avoiding selection bias is an important requirement for a valid case control study. Matching is a process wherein controls are selected so that some baseline variables are equally distributed between cases and controls and their confounding effect is minimized. A study by Deorari et al (4) is an example of case control study. Authors selected babies with pulmonary haemorrhage (cases). A control group of neonates matched for weight and gestational age born immediately before or after the included case were selected. The details of cases and controls available with case records were retrospectively retrieved and compared between two groups. RDS, PDA, hyperbilirubinemia, sepsis and DIC were found more commonly in cases than controls. This indicates association between pulmonary haemorrhage and these entities and NOT causeeffect relationship.

Cohort study

Cohort is a group of subjects sharing some common factor. In cohort study, the investigator selects a cohort of subjects and collects data about exposure and follows up the cohort for appearance of outcome of interest. The outcome of interest may have already occurred in cohort as in retrospective cohort or may not have occurred as in prospective cohort study design. Irrespective of whether design is prospective or retrospective, the measurement and data collection are always prospective. One example of cohort study is a study by Paul VK et al (5). They studied a cohort of women delivering after 26 weeks of gestation. They noted incidence of histological chorioamnionitis in this cohort. On comparing group with chorioamnionitis and group without, they noted a trend of increased incidence of preterm labor in presence of chorioamnionitis.

Interventional studies

The characteristic feature of interventional trials is that the investigator controls the exposure and follows up for effect. Such human studies are called clinical trials. Clinical trials may be performed with controls or without controls. Controlled trials are the common designs used in studies and have the advantage of minimizing errors. The gold standard experimental design is randomized controlled trial.

Randomized controlled trial (RCT)

Randomization is a process wherein each participant of a trial will have equal opportunity of being allocated to intervention group or to control group and such allocation cannot be predicted in advance. The advantage of randomization lies in its ability to give control group which is comparable to intervention group with respect to both known and unknown confounders.

An RCT can have parallel design or a cross over design. A parallel design is one wherein both intervention group and control group are concurrently followed up and outcome is measured. In cross over trial, intervention and control groups are mutually interchanged. Each treatment period in cross over trial is separated by washout period so that effect of prior treatment doesn’t persist.

Randomization can be done by various methods such as simple randomization, blocked randomization, stratified randomization, systematic randomization and cluster randomization. Blinding is a technique wherein the identity of the group i.e. intervention group or control group is concealed. When only participants are blinded it is called single blinded. If both participants and investigator are blinded it is called double blinded trial. Blinding reduces bias.

A recent example of a randomized control trial is on analgesic effect of expressed breast milk in procedural pain in term neonates (6). The design of the trial was randomized, placebocontrolled, double-blind one. 40 babies were enrolled in cases and 41 in control group. The intervention was feeding breast milk or distilled water before venepuncture. Outcomes measured included physiological and behavioral parameters. After comparison of two groups, authors noted that feeding 5 ml of EBM before venepuncture is effective in reducing symptoms due to pain in term neonates.

Systematic analysis and Meta analysis

All the study designs explained so far deal with human participants. A systematic review compares results of other studies. A systematic review tries to find all the relevant studies (published and unpublished), assess each study for the quality of its design and execution and combines the findings from individual studies in an unbiased manner. In this way they aim to present a balanced and impartial summary of the existing research evidence. Frequently, such systematic reviews provide a quantitativeestimate of net benefit aggregated over all the included studies. Such an approach is termed a meta-analysis. A Meta analysis of RCTs provides a strong evidence for clinical practice. Gottstein R (7) conducted a systematic review of use of intravenous immunoglobulin in hemolytic disease of newborn. They found 3 RCTs fulfilling inclusion criteria for the review. They combined the results of all three studies and come to the conclusion that need for exchange transfusion is reduced. Cochrane library (8) is an excellent source of systematic reviews and meta analysis in neonatology. The various study designs described so far form the basic types and other variations and combinations are also possible. For designing studies using other or more complicated designs, statisticians help should be sought for.

Choice of study design

A typical sequence for studying a topic begins with relatively easy and open ended observational studies that are followed or accompanied by analytical studies that analyze associations in order to discover cause–effect relationship. Experimental designs are used later as they are more difficult and answer focused questions.

Fig 1Comparison of case control, cohort and cross sectional study

Fig 2Parallel design

Fig 3Cross over design

Study Designs