In this talk, we will look at the different study design and types of data that are available in use in a epidemiology, to evaluate the impact of a screening programs. These elements are important in order to understand and to be able to quickly access the abundant literature which is available about screening evaluations. There are essentially three study design which have been used. The randomized controlled trials, the observational studies, which group the trend analysis of trend studies and case control studies. And the third is the modeling studies. So modeling studies are generally performed before a screening program or at the early stage to predict what would be the outcome. But not really directly use to assess, to evaluate a screening program, so I will not cover modeling studies. So let's look first at randomized control trial. It is the highest level of evidence we can have in medical studies. Everything is made to minimize the source of biases and to maximize the comparability between the screened and unscreened group. For the same recruiting methods, it is same criteria for include or exclude patients, The follow up approach is the same applied in ascertainment to the cause of the disease and to the test as well. So it is the best measure we have got available to assess a screening intervention. But the ideal design in fact does not exist and even random trials do have some weaknesses. One aspect is that the setting is quite different from the real world from the service screening program. Randomized control trial often call an ideal setting. So for instance the basis of professionals conducting screening is much wider, the proficiency, the skills also differ, so random control may not reflect real world situations. Also, the goal of randomized control trials is to compare invited and not invited screening which is not proficiently screened versus not being screened. So it does not measure itself the performance of the screening test. Another aspect is if the randomized control trials was conducted many years ago they could have been some subsequent changes for instance in the performance of the test available, the investigations we can perform or improvement in treatment, which cannot be reflected by all the randomized control trials. Finally, they always are the potential for residual bias. Some contamination effect, people were not supposed to be screened which were in the screened group and vice versa. And the balance of risk factors of the disease can not be ideally balanced even with randomization process. A second type of study which is used to evaluate impact of screening are case-control studies. It's actually probably the most commonly used design to evaluate several screening programs. So the basic principle of the case-control studies, if screening works, then people who experience the relevant clinical outcome should have a lesser history of screening than people who do not experience this outcome. In other terms, if the outcome is death which is often the case for instance in cancer screening, people if compared to cases people who died from the disease with control but people who are alive at the time of deaths of the cases. People who die from disease are expected to have been less often or not screened compared to the controls. These actually measure the efficacy of screening because it compares a group being screened, With a group which has not being screened. It does answer the simple question the patients may ask to his or her doctor. You, doctor may be recommending screening to patients and the patient may be asking « but doctor what protection can I expect from a regular screening? » It's quite different than the public health perspective looking at what is different in who should be invited versus not invited to screening. You're comparing people being screened versus people being unscreened, and this type of design allows some causal inferences about intervention. And we can, if we find an effect, say that the effect is related to screening. The major limitations of control-case studies, which is probably the major reason why case-control studies have often been neglected in review of the evidence about the effective of screenings is the self-selection bias. It is not randomized, people can choose, select, if they want to be screened or not and the big difference between these group in the risk of dying from the disease. They are a whole other means of correcting for this bias if we have some informations about the risk of disease among the unscreened populations. A difficult issue with case-control studies is often the lack of a true control population, especially the case if the whole population is invited for screening, who are you going to compare it with? And according to which control population you use it can lead to some different results. So it is one of the reason why case-control studies are sometimes not used when reviewing the evidence about screening because it tends to overestimate the benefit of the screening interventions. The third type of studies which have been used as trend studies or sometimes called ecological studies, when we're looking based on something simple such as vital statistics available. What are the trends of a disease? It is a very intuitive approach about screening evaluations. If screening works we should see an impact on the incidence of the mortality of the disease. So although trend studies are intuitive and very appealing by themselves, they carry a lot of weaknesses. The first and probably the main weakness is they do not allow causal inference. We cannot make conclusions based on aggregate data on specific individuals. Let's take an example outside the field of health and screening. You look at a school, different classes and there will be among these classes one who got the highest mark in mathematics. So you cannot conclude upon look at this pupil, he’s in this class but he must be a genius in math. Or look at his teacher, he must be excellent because He’s teaching the class with the highest average mark on mathematics. You cannot do this kind of inference. It is the same with screening, when you look at aggregate data at the population level you cannot make inference on individuals because you do not know if that particular individual has been screened or not. There is a great potential in trend studies to dilute the effect of screening, actually to underestimate its benefit. You may attribute some outcomes unrelated to the screening period. People may be dying in the screening period from the disease you screened, but the diagnoses was made before you started the screening program. It will also take many years for an effective screening to emerge. It can take several years before the full implementation of screening program occurs and these effect must be taken to account to look at the right time windows looking for also at the right age group to measure the potential of the screening programs. There're also other potential confounding effects. They may be a greater awareness of a disease, after screening started improving treatment, it can be different causes such as better disease registration, better recording of cause of death and all these aspects may occur at a macro or micro level, area level cannot be taken into account with trend studies. So although it’s a very appealing tool, very easy to use, it is probably the tool by itself which has the most weakness and cannot alone be used to evaluate the impact of a screening intervention. The last issue with trend studies, if we look at result at a population level, that we do not have any control groups which is readily available and you also need to have aggregated data, vital statistics before screening started so you could see the background effect and see what is the reference level, whether for the incidence or the mortality for the disease. So in conclusion there is no ideal study design to evaluate the impact, whether it is incidence or mortality from screening. It is always a good approach to use different study designs to evaluate a screening program.
