8 Operational Research
Before undertaking operational research, it is important to have a plan. You need to decide:
- What question you are trying to answer (your research question)
- What your aim is in doing the research
- Where you will get your data from and how
- What methods you will use to analyse your data
- How you plan to use or share the findings of your research
- How you will ensure that participants are not harmed by your research (e.g. how you will keep patient information private)
Descriptive epidemiology
Descriptive epidemiology is describing the what, when, where, and who of health determinants or conditions in a population. What (clinical features or health condition); who (person); where (place); when (time).
Descriptive epidemiology can be used for surveillance data, field projects (e.g. descriptive cross-sectional studies) and in outbreak investigations. With descriptive epidemiology we can start to form hypotheses about a possible source of an illness during an outbreak investigation.
Analytic epidemiology
Analytic epidemiology uses statistical methods and comparison groups to assess how much evidence there is in our data to make a conclusion about what is happening. It looks to identify and measure associations, test hypotheses, and identify causes. It uses a comparison group (or groups) to confirm or reject an association between risk factors and illness between the groups. Case-control studies and cohort studies are used as analytical studies to test hypotheses about a possible source of illness during an outbreak investigation.
Association
An association means that there is evidence from our data that two or more events are related to each other. In epidemiology, the events we are typically looking at are exposure and outcome (disease).
Experimental study
A study that involves an intervention and then analyses what happens. In an experimental study the investigator specifies the exposure category for each individual (clinical trial) or community (community trial), then follows the individuals or community to detect the effects of the exposure. Experimental studies can be used in field epidemiology but they are not common because they tend to be long, costly and involve complex ethical issues.
Observational study
Observational studies are the most common type of analytical study used in epidemiology. They involve the investigator observing what has happened or is happening in a group. Changes or differences in one characteristic are studied in relation to changes or differences in others, without the intervention of the investigator.
Hypothesis
A hypothesis is an idea that can explain something. It is based on known information but has not yet been tested to see if it is true. A hypothesis is tested through analytic studies to see if it is supported by evidence.
Null hypothesis
The null hypothesis is always that there is no difference or association between the items under study (could be groups, populations, variables etc). The null hypothesis assumes any observed difference is due to chance and the difference would likely not be observed if we collected the data again.
Alternative hypothesis
The alternative hypothesis goes against the null hypothesis and predicts that there is a significant difference or association between the items under study. We usually think of our alternative
hypothesis first because it is our idea about what is happening. The null hypothesis is then simply that our idea is not true.
Example:
An epidemiologist is studying the relationship between malaria incidence and annual average rainfall. Their hypothesis is that malaria incidence is higher in locations with high rainfall compared with locations with low rainfall.
The null hypothesis would be that there is no significant difference in the incidence of malaria between areas with high rainfall and areas with low rainfall.
The alternative hypothesis would be that the incidence of malaria is significantly higher in areas with high rainfall compared with areas with low rainfall.
The epidemiologist would then design an appropriate study to test these hypotheses.
A descriptive study can help us come up with hypotheses, whereas an analytic study can test them.
Case reports
A type of descriptive epidemiological study that reports individual cases of unusual diseases or outcomes.
Often clinical reports are of rare diseases or rare outcomes. Often used to describe first case of an emerging disease.
Example:
When Zika virus began spreading rapidly across Brazil in 2015, case reports were written about individual cases or, in this case, twins who had mirocephaly and confirmed Zika virus (Santos et al (2017) DOI: 10.4269/ajtmh.16-1021). In this article, the authors describe in detail the clinical characteristics of these two babies.
Case series
A descriptive epidemiological study that is similar to case reports but describes multiple cases. These cases are usually related to each other in some way, such as coming from the same location, population group or time period.
Example:
In the Zika virus outbreak in 2015, there were then enough cases detected to produce a case series (Franca et al (2016) DOI: 10.1016/S0140-6736(16)30902-3). This article describes the clinical characteristics of the first 1501 live births in Brazil.
Cross-sectional study
A cross-sectional study looks at data from a population at one point in time; it is a snapshot of what is happening in this population at this point in time. It is often called a ‘prevalence study’ because it is mostly used to estimate prevalence of a health condition in a population. A cross-sectional study can be either a descriptive study or an analytical (observational) study.
The exposure and outcome are measured at the same time, so it cannot examine cause and effect; you cannot use a cross-sectional study to see what causes a disease. Participants are selected as a sample from the population you want to study and data is collected from each participant on the variables you want to estimate. The researcher only measures the data variables present in the population; they do not do anything to try to change these variables.
Advantages and disadvantages of cross-sectional studies include:
Advantages |
Disadvantages |
|---|---|
|
|
Example:
A cross-sectional study looked at 120 infants who had been exposed to Zika virus in utero to see the proportion of these infants who also had heart issues (Orofino et al (2018) DOI: 10.1371/journal.pntd.0006362).
By looking at the variables – heart issues and in utero exposure to Zika virus – the study team were able to measure the frequency of heart issues amongst those children with in utero exposure to the virus. They determined that there was higher prevalence of major heart issues amongst those infants with in utero exposure.
The study did not attempt to change the variables, or to conduct an intervention or to examine whether in utero exposure to Zika virus caused heart issues.
The study shows a snapshot in time of the outcome and exposure; the prevalence of major heart issues amongst those infants exposed to Zika virus in utero.
Case control study
A type of observational analytic study. Participants are selected for the study based on having the disease being studied (“case”) or not having the disease (“control”). Data are collected on exposures or risk factors that you think might be associated with the disease and then compared between cases and controls to determine if there is an association between exposure and outcome.
If you wanted to investigate a foodborne outbreak in the community, you would select your cases (people who got sick with gastroenteritis, e.g., Diarrhoea and vomiting) and your controls (people who did not get sick with gastroenteritis) and investigate all of the different foods that each group of people had eaten during the same exposure period. In doing so, you can check if there was an association between eating a particular food/s and getting sick.
Advantages and disadvantages of case control studies include:
Advantages |
Disadvantages |
|---|---|
|
|
Cohort study
A type of observational analytic study. Participants are selected for the study based on whether or not they have a particular exposure or risk factor. The groups (exposed and unexposed) should be very similar (e.g. similar proportions of men and women) except for the exposure or risk factor you are investigating. Health outcomes that may be associated with that exposure or risk factor can then be analysed.
A cohort study can be prospective or retrospective.
A prospective cohort study is a study that selects participants based on their exposure and then follows them into the future to see what health outcomes they develop. A prospective cohort study may recruit participants when they are in their 20s, for example, and follow them throughout their whole lifetime to compare if the outcome (e.g., lung cancer) is different between the two groups (e.g., Smokers and non-smokers).
A retrospective cohort study is a study that uses data from the past, either from asking participants about the past or using data that was already collected for another reason (e.g. health records). You would first check if the data shows that they were exposed to the risk factor being studied and then divide them into exposed and unexposed group (e.g. whether they smoke or not). You then check if they currently have or previously had the health condition/s you are studying (e.g. lung cancer) and see if this differs between the exposed and unexposed groups. A retrospective cohort study is very similar to a case-control study, except the groups are divided based on their exposure to a risk factor instead of whether or not they have the health condition.
A retrospective cohort study can be used in outbreak investigations in populations that are well defined (e.g. all the attendees at an event, such as a wedding).
Advantages and disadvantages of cohort studies include:
Advantages |
Disadvantages |
|
|
The below diagram shows the direction in which the investigation goes for each study type.
Sample
Sometimes when you are researching a group of people, you have data for the whole group (e.g. a population census, or a small, local outbreak where you have data for all the people who were at risk). However, this is rare and usually you can only get data for a part of the group you are researching. In this case, you can select part of the group to participate in the study. This is called a sample.
A sample is studied to get evidence to estimate what is happening in the whole population. A sample may be selected using random or non-random methods, and the sample itself may be representative or non-representative of the larger population.
Example:
You are conducting a survey in a village with 1000 people. You only have one day to do the survey, so it is not possible to ask all 1000 people. Surveying 100 people is a sample of your overall population, but not necessarily representative of everyone’s responses.
A representative sample has characteristics that are similar to the whole population you are interested in. These characteristics may be related to age, sex, location, occupation, presence or absence of a risk factor or many other possible options. A non-representative sample can give you an idea of what may be happening in the population, but a representative sample gives you more confidence that what you find in your sample accurately reflects what is happening in the whole population.
Example:
You are interested in interviewing healthcare workers at a hospital. You know that there are 100 healthcare workers, and 70 of them are women. You may only have time and resources to interview ten healthcare workers. To make this sample more representative of your overall population (all healthcare workers at the hospital) you could invite seven women and three men to be interviewed, so that women make up 70% of the sample to match the proportion of women in the population.
Sampling methods
There are multiple ways to select a sample.
Below is a description of a few of the most common sampling methods used by field epidemiologists.
Random sampling
A sampling method whereby each person in the population has the same chance of being selected.
Example:
total people in the village. You could enter all the names into Excel and then randomly choose 100 people, for example, or you could choose every second house as you walk around the village. In this way, each person has the same chance of being selected to participate in the survey.
Simple random sample
Convenience sampling
A sampling method where the researcher selects people who are easy to contact. It is a simple and cheap way to collect data. The data are not necessarily representative of the population.
Example:
You want to understand challenges nurses face in clinical work during COVID-19 by interviewing 20 nurses. Whilst it would be ideal to explore the different challenges faced by nurses in different settings (e.g., Rural or urban, private or public facility), you have a short deadline and limited funding which means you cannot travel or pay for participants’ travel. Convenience sampling in this example would be to select 20 nurses that work at the health facility closest to you. The sample will not be representative of the whole nursing profession but will give insights into some of the challenges faced by nurses working during COVID-19.
Convenience sample
Purposive sampling
A sampling method where the researcher selects their sample based on participant characteristics that best meet the purposes of the research. There must be clear criteria for inclusion. It is used commonly in qualitative research studies where the researcher wants to understand the experience of a particular event, or when the study population is very small (e.g., people with an extremely rare disease).
Example:
You want to understand why some patients with tuberculosis become lost-to-follow-up (LTFU). You purposefully select a number of LTFU patients to collect data on a range of experiences.
Purposive sample
Snowball sampling
A sampling method where the researcher recruits participants who were suggested by other (already recruited) participants. Snowball sampling is particularly useful if the population is hard to find because the characteristic of interest is rare, there is stigma related to the characteristic, or because there is no organised way of identifying people with the characteristic.
Example:
You want to understand how homeless women access reproductive health services in your city. Once you identify one woman who agrees to participate, she informs you of other women in the same situation and how to contact them.
Snowball sample
Bias
Bias is when the way that we collect, analyse, and interpret data that leads to conclusions that may not be true. Bias means that we are limited in what we can do with the results of a study.
It is very common for studies to have some error; error can be random or systematic. Random error occurs due to chance and should not affect your study results too much. Systematic error occurs when the same type of error repeatedly occurs leading to bias and potentially an incorrect study conclusion.
Epidemiological studies always have at least some bias. Bias needs to be recognised and addressed in study design, planning, and implementation as it is difficult to fix or control once data have been collected.
The are many types bias; this section will focus on some of the main types that field epidemiologists need to consider in operational work and project design and implementation.
Selection bias
Selection bias is a result of factors that affect who participates in the study, either because of errors in how participants were selected or because of who chooses to participate and who doesn’t. The result is that there is a real difference in the those who participate and those who do not participate that affects the study’s findings. The relationship between exposure and outcome observed in the sample is not representative of the relationship in the whole population.
The following types of selection bias are common in field epidemiology and should be considered in study design and implementation:
Sampling bias
Sampling bias occurs when the study participants are selected in a non-random way. This means that some members of a group are more likely to be selected to participate than others. Some groups in the population may be excluded from selection, which can lead to a real difference between participants and non-participants. This means that results are not able to be generalised to the broader population; they are applicable only to people with the same characteristics as the selected sample.
To address sampling bias, we ideally select study participants randomly. In this way, each person in the population has an equal opportunity of being selected to participate, and we reduce sampling bias.
In some circumstances (e.g., resource or time limited situations or when the population is small), we may choose a non- random sampling process, such as convenience sampling, for a project. Convenience sampling increases the likelihood of sampling bias because we are intentionally selecting people based on a known characteristic that may be different to non-participants. Convenience sampling is commonly used in epidemiological studies; it is important to understand and be able to describe the limitations of this sampling strategy in terms of how convenience sampling may impact the generalisability of the findings. The study findings can still be useful, we just can’t assume that they are true for the population so should be careful what actions we take based on the findings.
Non-response bias
Non-response bias is where the people who do not respond to study invitation or who drop out are different from those who agree to participate.
Example:
In a project wanting to understand the outbreak investigation experiences of FETP graduates, people living in remote areas with limited or no internet network may be more likely to not respond to project invitations sent by email than those based in urban areas with good internet network. In this case, the outbreak investigation experiences may be very different between urban and remote FETP graduates, and as such, the findings will not reflect the true range of experiences of graduates.
Volunteer bias
Volunteer bias is when those who volunteer to participate are different to those people who do not volunteer to participate. Participants are more likely to participate in studies and projects where they are interested in the topic being studied. As such, they may not reflect the population as a whole.
Example:
In a project aiming to evaluate the impact of a community exercise program, people who are already committed to exercising regularly may be more likely to participate than those who do not already have regular exercise practice. This difference may influence the results and our conclusions as those who volunteer for the study are meaningfully different to those who do not.
Information bias
Information bias occurs when exposure or outcome data are collected in a different way between study groups. This can lead to an incorrect estimate of the association between exposure and outcome. Some common types of information bias are explained below.
Recall bias
Recall bias occurs when one study group tends to remember information differently to the other study group. Information provided by one group may be less accurate than that of the other group. This is a common issue in retrospective studies where people are self-reporting information.
Example:
A case-control study where cases are selected based on having the outcome (e.g., a disease). Cases may have thought extensively about their disease, risk factors, and the things they did before they got sick. Cases may therefore be more likely to recall “exposures” than the controls, who do not have the disease/outcome.
Social desirability bias
Social desirability bias occurs when respondents answer a survey or questionnaire in a way they think will be seen as better by others.
Example:
In a study exploring the association between exercise (exposure) and a health event (outcome), participants are asked about their level of exercise. If exercise is viewed favourably in their community they may say they exercise three times per week when it is really only one.
Interviewer bias
Interviewer bias occurs when the person conducting interviews in a survey or study influences the responses given by respondents. This can be an issue if questionnaires are not standardised or when the interviewer knows the outcome or exposure status of an individual, causing them to ask leading questions or change the way the questions are asked.
Confounding
Confounding occurs when a third factor changes the association between the disease (or outcome) and an exposure (or independent variable).
Confounding occurs when the third factor is associated with both the exposure in question and the disease (outcome). Confounding can be minimised in how a study is designed and analysed.
Example:
Epidemiologists want to study the association between exercise and heart disease. Age could be a potential confounder because it is associated with both exercise and heart disease. Age is a known risk factor for heart disease. Also, a person’s age could influence their ability to participate in physical exercise. It is important the researchers account for age in their analysis – either analysing the data by age group (stratifying) or using statistical methods to adjust for age. If they don’t do this, they might find misleading results which have been confounded by age.
