Scribd
Upload
6 views5 pages

Study Design

Study Design

Uploaded by

Kristina Gärtner Kralik
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
6 views5 pages

Study Design

Study Design

Uploaded by

Kristina Gärtner Kralik
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 5
Metical Statistics At A GLANCE sensinn Mea aes UN ae TT @ Study design I Study design is vitally important as poorly designed studies may ive misleading results. Lange amounts of data from a poor study will not compensate for problems ints design. Inthis chapter and in Chapter 13 we discuss some ofthe main aspects of study design. In Chapters 14-16 we discuss specific types of study: clinical til cohort studies and case-control studies. ‘The aims of any study shouldbe clearly stated atthe outset. We may wish to estimate a parameter inthe population (such as the risk of some event (Chapter 15), to consider associations between a panicular aetiological factor and an outcome of interest or teval- tate the effect ofan intervention (suc as a new treatment). There ‘may be a numberof possible designs for any such study. The ult- mate choice of design will depend not only on the sims, but on the resources available and ethical consideration (see Table 121) Experimental or observational studies + Experimental studies involve the investigator intervening in some ‘way to affect the outcome, The clinical trial (Chapter 14) is an Table 121 Study designs. example of an experimental study in which the investigator ito- «ces some form of teatmen. Other examples include animal staes or laboratory sais that are cared out under experimental conde tions, Experimental studies provide the mostconvincing evidence for any hypothesis asi is generally possible to enol fr actors that ‘may affect the outcome, However, these stuliesare not alway eas ‘be og if they involve humans or animals, may be woethica. + Observational studies, for example cohort (Chapter 18) oF case-control (Chapter 16 staies are those in which the investiga- tor does nothing to affect the ourcome, but simply observes what happens. These stadies may provide poorer information then experimental studies because itis often imposible to contr forall factors that affect the outcome. However, in some situations, they tay be the onl types of study that are helpful or possible, Ep- demiological studies, which assess the relationship between factor of interest and disease in the population, are observational ‘sion in Actionin _preseattime Action in ‘Typeofstudy —Tining Form pastime (starting point) future time ‘Typical uses Cross-sectional Cross Observational] [eater] * Prevalence eimates| seciional al + Reference ranges and diagnostic ial test ia + Curent ahh status ofa group y v yv Repested Cross Observational a Gea) [eater] * hangs overtime ‘rose-setional sectional = information Cohort Longitudinal Observational Dain cohort + Prognosis and natural history (Chapter 18) (prospective) ‘and yt peo omece asses ik actors mise CCase-consol Longitudinal Observational seer cneane + Aetiology (aricuary forrare (Chapter 6) etrospective) Moe | ned endconrols senses) factors |#—|_Ge. outcome) Experiment Longitudinal Experimental a tae * Cn a oases egy eee fatow + Trial to assess preventative yn Study design | ‘measure, e.g. large scale vaccine tal + Laboratory exgeriment Assessing causality in observational studies Although the most convincing evidence for the causal role of a factorin disease usually comes from experimental studies, informa- tion from observational studies may be used provided it meets a umber of criteria. The most well known criteria for assessing cau sation were proposed by Hill, + The cause must precede the effect. + The association should be plausible, ie. the results should be bio- logically sensible + There should be consistent results from a number of studies. ‘+ The association between the cause and the effect should be strong. + There should be a dose-response relationship with the effect, i. higher levels of the effect should lead to more severe disease or more rapid disease onset. + Removing the factor of interest should reduce the isk of disease. Cross-sectional or longitudinal studies + Cross-sectional studies are carried out at a single point in time. Examples include surveys and censuses ofthe population. They are particularly suitable for estimating the point prevalence of acondi- tion in the population, [Number with the disease ata single time poi Total number studied atthe same time point Point prevalence: ‘As we do not know when the events occurred prior to the study, wwe can only say that there is an association between the factor of interest and disease, and not that the factor is likely to have eaused disease. Furthermore, we cannot estimate the incidence of the disease, ic. the rate of new events in a particular period (Chapter 31), In addition, because cross-sectional studies are only cartied out ‘atone point in time, we cannot consider trends over time. However, these studies are generally quick and cheap to perform. + Repeated cross-sectional studies may be carried out at different time points to assess trends over time. However, as these studies involve different groups of individuals at each time point, itcan be "Hil, AB, (1965) The envionment and disease: association or causation? Pro ceedings ofthe Reval Society of Medicine, 8,295. difficlt co assess whether apparent changes overtime simply reflect differences in the groups of individuals studied. + Longitudinal studies follow a sample of individuals over time, ‘They are usually prospective in that individuals are followed for~ ‘wards from some point in time (Chapter 15). Sometimes retrospec~ tive studies, in which individuals ae selected and factors that have ‘occurred in their past are identified (Chapter 16) are also perceived as longitudinal. Longitudinal studies generally take longer to carry ‘out than cross-sectional studies, thus requiring more resources, and, if they rely on patient memory or medical records, may be subject to bias (explained atthe end of this chapter) Experimental studies are generally prospective as they consider ‘the impact ofan intervention on an outcome that will happen inthe future, However, observational studies may be either prospective or retrospective. Controls ‘The use of comparison group, or control group, is essential when designing a study and interpreting any research findings, For example, when assessing the causal role of a particular factor for adisease, the risk of disease should be considered both in those ‘who are exposed and in those who are unexposed to the factor of interest (Chapters 15 and 16). See also “Treatment comparisons" in Chapter 14 Bias When there is a systematic difference between the results from a study and the true state of affairs, bins is stid to have occured. ‘Types of bias include: + Observer bias—one observer consistently under-or over-reports ‘particular variable; + Confounding bias — where a spurious association arses due to @ failure to adjust fully for factors related to both the risk factor and ‘outcome (see Chapter 34); + Selection bias patients selected for inclusion into a study are not representative of the population to which the results will be applied “+ Information bias— measurements are incorrectly recorded in a systematic manner; and + Publication bias—a tendency to publish only those papers that report positive or topical results. (Other biases may, for example, be due to recall (Chapter 16), healthy entrant effect (Chapter 15), assessment (Chapter 14) and allocation (Chapter 14). Study design | Study design 31 ‘Variation in data may be caused by known factors, messurement ‘errors’, or may be unexplainable random variation, We measure the impact of variation inthe data on the estimation ofa population parameter by using the standard error (Chapter 10). When the mea- ‘surement ofa Variable is subject to considerable variation estimates relating to that variable will be imprecise, with large standard ‘errors. Clearly it is desirable to reduce the impact of variation as far as possible, and thereby increase the precision of our estimates “There are various ways in which we can do this. Replication ur estimates are more precise if we take replicates (e.g. Wo oF three measorements ofa given variable fr every individual on each fecasion), However, as replicate measurements are not indepen- deni, we must take care when analysing these data. A simple approach isto use the mean of each set of replicates in dhe analysis in place of the original measurements. Alternatively, we can use rethods that specifically deal with replicated measurements (see Chapters 41 and 42) pt ‘The choice of an appropriate size for a study isa crucial aspect of study design. With an increased sample size, the standard errarof sn estimate will be reduced, leading to increased precision and study power (Chapter 18). Sample size calculations (Chapter 36) should be carried out before starting the study. Partiqulsr study desig Modifications of simple study designs can lead to more precise est- mates. Essentially we are comparing the effect of one oF more ‘treatments’ on experimental units, The experimental unit isthe sinallest group of ‘individuals’ which can be regarded as indepen- dent forthe purposes of analysis, for example, an individual patent ‘volume of blood oF skin patch. If experimental units are assigned sandomly (i.e. by chance) to treatments (Chapter 14) and there are no other refinements to the design, en we have a complete ran- ‘domized design, Although this design i straightforward to analyse, itis inefficient if there is substantial variation between the experi- ‘mental units, In this situation, we can incorporate blocking and/or use a cross-over design to reduce the impact of this variation. Blocking It is often possible to group experimental units that share similar characterises into a homogeneous block or stratum (e.g. the blocks may represent different age groups). The variation between units in a block is less than thst between units in different blocks. ‘The individuals within each block are randomly assigned to teat ments; we compare treatments within each block rather than ‘making an overall comparison between the individuals in different blocks, We can therefore assess the effects of treatment more pre- cisely than if there was no blocking. Parallel versus cross-over designs (Fig. 13.1) Generally, we make comparisons between individuals in different 82. Stay Study design I ‘groups. For example, most clinical trials (Chapter 14) are parallel vials, in which each patient receives one ofthe two (or occasionally ‘more) treatments that ar being compared, i.e. they result in benween-individual comparisons. ‘Because there is usualy less variation in a measurement within an individual than between different individuals (Chapter 6), ia some situations it may be preferable to consider using each individ- tual as his/her own control. These within-individaal comparisons provide more precise comparisons than those from between-indi- ‘vidual designs, and fewer individusls are required for the study 10 ‘achieve the same level of precision. In a clinical ial seting, the ‘eross-over design! is an example of a within-individual compari- son; if there are two treatments, every individual gets each tteat- ‘ment, one after the otberin a random order to eliminate any effect of calendar time. The treatment periods are separated by a washout period, which allows any residual effects (carry-over) of the previous treatment to dissipate. We analyse the difference in the responses on the 1wo treatments foreach individual. This design can only be used when the treatment temporarily alleviates ‘symptoms rather than provides a cure, andthe response time is not prolonged, Factorial experiments ‘When we are intrested in more than one factor, separate studies that assess the effect of varying one factor ata time may be ineffi- cient and costly. Factorial designs allow the simultaneous analysis ‘of any number of factors of interest. The simplest design, 9 2x 2 factorial experiment, considers two factors (For example, (0 dif= ferent treatments), each at to levels (e.g. either active or inactive treatment). As an example, consider the US Physicians Health study? designed to assess the importance of aspirin and beta carotene in preventing heart disease and cancer, A 2 x 2 factorial

You might also like