So cross-sectional design is used to describe the magnitude of a condition or a disease. Many times, data for this study design are collected using the survey or questionnaires. Like case report, a survey is also categorised as observational and descriptive research design.
However, the study unit is targeting a much bigger population than case studies. Survey is a good way to collect large amounts of public health data quickly and survey grew out of social science such as marketing, economics, or politics dealing with human behaviours. And because much of public health depends on behaviours, survey became a useful tool to measure past and present behaviours in public health research.
So the main feature of cross-sectional design is that it is descriptive and so because this is descriptive it can be useful as a starting point to generate more hypothesis for further investigation about the development of disease. And the goals of the cross-sectional design is the are to estimate the magnitude of disease or health condition in a defined population by providing the prevalence information. Another goal of cross-sectional study design is to assess the need for health service and to establish counter-measures against the condition to relieve the burden of health care.
There are a number of population-based cross-sectional studies on blindness and eye diseases. For example the Irish glaucoma survey in 1993 was a successful cross-sectional study that provided prevalence of glaucoma in people over 50 years old. Another cross-sectional study, North London eye study in 1998 measured prevalence of serious eye diseases such as cataracts or age-related macular degeneration in people over 65 years old.
Every year, a number of non-profit organisations such as WHO or other NGOs publish annual reports on global and original estimates of the prevalence of visual impairments and other major blinding eye conditions. Results from such surveys are used to improve the eye care services or health policy. So in order to understand the epidemiology of visual impairment, we need to use an agreed definition of visual acuity so that we can compare the result between the studies and measure impact more accurately.
The WHO categories are based on the visual acuity of the better eye and presenting vision. Pinhole is not used when using this method and presenting visual acuity provides an understanding of refractive need and is important because especially for the epidemiological surveys and planning. So WHO has agreed on the use of four different levels of visual function as described in the international classification of disease-10.
So these four levels are mild or no visual impairment, moderate visual impairments, severe visual impairment, and blindness. So each of these four levels has a clear criterion based on the visual acuity measurement. So according to this classification, moderate visual impairment, moderate and severe visual impairment are grouped under the term low vision whereas the low vision is taken together with the blindness and then that represents all visual impairment.
So this is the table of visual impairment categories based on the international classification of a disease tenth revision by WHO, which I showed you before when I talked about the levels of measurement. So the WHO's four levels of a visual function are categorised and measured as follows. So category 0 is mild or no visual impairment and in this category, vision is equal to or better than 6 over 18 using Snellen fraction and category 1 is moderate visual impairment where the vision is worse than 6 over 18 but equal to or better than 6 over 60.
and category 2 is severe visual impairment and in this category vision is worse than 6 over 60 but equal to or better than 3 over 60. And then the final category, the category 3, is blindness and here the vision is worse than 3 over 60. So it is important to remember that visual status categories are based on visual acuity measuring the better eye and blindness is also used if the visual field is less than 10 degrees from a point of fixation.
In 2010, the WHO reported that there were 285 million visually impaired people in the world based on the prevalence figures calculated from cross-sectional studies. So here is one of the visual summaries of global estimation of blindness by different age groups from the same report using their categorisation of visual function. So as illustrated here when the overall prevalence of blindness is broken down by age, blindness, whatever the causes are, affects mainly the older section of the population, which are these two groups right.
But then when it does, it hits the age groups really hard because until like age 44 and the prevalence is pretty low but then after that age group, the prevalence of blindness that goes up like whooping 20 times in this age group and it just goes up exponentially. So survey or questionnaires are quick and easy way to do a cross-sectional study. But the topics ideal for survey are those that can be accurately measured by responses from the respondents such as past and current health behaviours.
However, measuring disease prevalence through survey will be very likely to be biased without a proper diagnosis from a trained health professional. So from that sense, one of the representative example of survey used in the public health is quality of life assessment. So this is a fast-growing research area in healthcare using survey design.
In old days, most medical decisions are exclusively dominated by the healthcare professional's authority. However, healthcare system now has evolved from a disease based to a patient-centred model that is based on the idea that what healthcare professionals find useful and working does not always go hand-in-hand with well-being and the functional capacity of patients. So what's shown here is a visual functioning questionnaire developed by the national eye institute in America that was originally developed to assess the patient's quality of life before and after treatment or surgery in the eye care but it can be used in a broader context.
So in conducting cross-sectional study using survey, you need to think about who will be your subject for your research so this process requires a clear consideration of the target population and the population to which the main results of the study will be extrapolated. For example, if you want to study the distribution of visual impairment in Scotland then your target population will be the entire population living in Scotland but for practical reasons, it would be impossible to include all the population across Scotland. So then you need to identify a suitable source population, for example, you might decide to conduct the study in one of the major cities like Glasgow.
And if this is still too big, then the representative sample needs to be selected from the source population using a sampling scheme that we will talk about shortly. So from this sample the final study participants will be determined because not everyone in the selected sample will end up participating in the study. Some subjects will refuse to participate despite all reasonable efforts.
On the other hand, others will have moved out of the area or even dead at the moment. So the final participants are usually a subset of the initial sample who are eligible and willing. So a sampling scheme is a detailed description of what data will be obtained and how this will be done.
So here we will briefly talk about three major sampling schemes among others. So first, we have a simple random sampling scheme which is the most basic and straightforward sampling scheme. So this is a method of selecting n number of samples from a population of size N such that every possible sample of size n has equal chance of being drawn.
So this type of sampling scheme provides the most precise results with the smallest sampling error for a given sample size when other things are equal. And one of the example studies used this simple sampling scheme is the Irish glaucoma survey that I mentioned previously in 1993. They actually successfully applied this sampling scheme to measure the prevalence of glaucoma in the older population.
So here is kind of a cartoon illustration how simple random sampling is done. So here we have a group of 30 and so that is our size of population and we want to select random sample of 15. So here it is assumed that each individual have an equal chance of being selected.
So before we select the sample, we're going to assign whole numbers in sequence to each member like this right and you draw 15 random numbers from 30 and pick the corresponding members to have a random sample. So hopefully this kind of reminds you of something you've done already which so because this is basically what you did last week to sample your own data for the data analysis task. Next up, we have a stratified random sampling scheme and in this scheme the population of N units So in this scheme, you first divide the population of interest into several mutually exclusive subgroups called strata based on the variables of interest such as age or gender which are typically unmodifiable risk factors so for example if you want to know the differences in refractive errors of a population between the different age groups let's say less than 20 years old, 20 to 40, and over 40 then you might want to stratify the population into the corresponding age groups then do the simple random sampling so that relatively equal proportion of each age group can be selected.
So again in this cartoon illustration we will select 15 random samples out of 30. So as you did in the simple random sampling you're going to assign each individual with the sequence of numbers. But before you do the sampling you want to identify strata because you're interested in the gender difference so you want to maintain a relatively equal proportion of gender that is distributed in the population.
So in that case you stratify the population by gender and in this example our population is stratified by 10 males and 20 females. and then now you can draw 50 random numbers from each stratum and pick the respective numbers as the number of samples we wanted to sample from the beginning was 50 percent of the populations anyway. Now you can use the simple random sampling scheme.
that is male and female. and finally the cluster sampling. So in real-world situations, simple random sampling is not cost-effective especially when you are carrying out a study based on a large population.
For example to measure reflective areas of the Scotland population and you want to use a sample and select it by simple random sampling then the study participants you have to cover will be randomly dispersed over a quite large area which will be quite costly and expensive. So instead you can divide the greater Glasgow area into a smaller grid and pick n number of geographical areas through simple random sampling then pick the study samples from the chosen area using again the simple random sampling. That way, you know where your study samples are clustered and that way you will be able to save time and resource in collecting data.
There are a number of studies have used this cluster sampling such as a Nepal blindness survey in 1981 or Baltimore eye survey in 1985 in America. So in cluster sampling, before you do the sampling you want to divide the population into several geographical clusters like this as an example and then you assign each cluster with the sequence of numbers and you generate n random numbers from the N number of clusters you have in this case the big N is 12 and then the small n, in this case, what one two three four five right so basically you pick out five random clusters out of 12 clusters. So once the clusters are sampled then you do not do the further sampling so you're going to just take the whole population from each cluster so that's how you do the cluster sampling.
So here are the good things and bad things about the study design. So among other things, cross-sectional design can provide prevalence information which is a baseline rate of a disease or diseases within a population which in turn provides valuable information about the current status of a disease or condition in that population. So this type of research is relatively cheap and quick to complete compared to the other epidemiological research design.
Of course, it is more expensive and takes more time compared to the case report and but this is especially so when you have access to routinely collected data from healthcare providers such as NHS or hospitals so if you do have access to those data then it is much easier to carry out a large-scale cross-sectional study with a minimum cost which is a major advantage over other forms of epidemiological research design. On the other hand one of the major drawbacks of this design is that it is not appropriate to test causal relationship between the risk factors and prevalence. Because both exposures and diseases are studied at the same time, the time sequence between risk factors and disease cannot be established, which is known as a temporality.
We're going to just talk about this later on and one of the major assumptions behind the disease occurrence is that the risk factors or the cause always come before the disease in time. So that is the temporality and you cannot really establish this temporality with a cross-sectional design because they are, exposures, the risk factors and diseases are studied at the same time. And finally, a cross-sectional study is also not adequate to study rare conditions.
because you need to survey a large sample, very large sample to obtain enough cases and cases of short duration have a smaller chance of being detected in a one-time cross-sectional survey than the cases of longer duration because people with short-lived diseases will either recover fast or die fast. So inevitably chronic diseases are typically over-represented in a cross-sectional survey.