Imagine an old woman walking along a path. The path is icy, slippery, and the city council has not cleared up the snow that fell the day before. The old lady, now about 80 years old, slipped on the icy pavement, fell and broke her hip. A few local people who gathered around her, helped her to an ambulance, and she was rushed to a nearby hospital. She told the doctor that ten years ago, she suffered from a transient ischaemic attack in her brain, and this left her a bit "wobbly". From this history, can you identify what was the cause of her hip fracture? Was it the icy pavement alone, was it the wobbliness? Was it her 10-year old ischaemic attack?Call to action: Can you list as many causal mechanisms as you can think of?What is meant by cause? Cause is defined as an event, or a circumstance that leads to another event and the causal variables must be present immediately preceding or precede by some time before the outcome can occur ....Ken Rothman (1993) has argued that our notions of causes are intuitive from the time we are born\cite{Rothman_1993}. A baby knows that if the baby cries, she is going to be nursed and given the feed. This is a notion of cause that encourages that causes and effects are one-on-one. But in health and healthcare, most causes are multifactorial and there are usually more than one cause to every outcome. Such a situation demands that we discuss different causes for the same outcome. We could draw up a number of pie charts if we knew the relative contributions of each of the causes of the outcome (we will learn how to calculate such pie slices). Each of these pie charts would indicate a set of components that together would define the sufficient conditions that if these causal constellations were to be found, then they would define the outcome. Sufficient causal models are those causal models where more than one cause can be account for a particular health outcome. These causes biologically interact. Each component of the causal model in the form of the slice of the pie (see Figure 1) is referred to as a component cause. A component cause that is present in most or all of the sufficient causal model and without which the outcome cannot occur is referred to as a necessary cause.Search for causes to explain health events is embedded in the history of health sciences. In the beginning it was based on the opinions of the thought leaders of the day, and referred to as "scholasticism" \cite{hankins2007humanism}. Scholasticism was replaced with inquiries on the nature of truth using inductive logic of John Stuart Mill. In inductive logic, the scientist would start with an observation of fact and from the fact, he would arrive at a larger theory that would explain the world. For example, to an inductive logician, the observation of an elderly lady falling on the icy pavement and fracturing her hip would lead to a theory that all people walking on icy path are prone to fracture their hip. The tenets of inductive logic was challenged in the nineteenth century by proponents who argued against the core assumption that the fact that formed the basis of inductive logic would not vary over time or place. This assumption may not be true, and therefore the proponents of a new logic (the school of deduction) argued that refutation of hypotheses would be a way to establish truth. David Hume was a proponent of this theory of skepticism to examine the hypotheses \cite{schmidt2010david}. In much of the nineteenth and the twentieth century, this form of logic was dominant. Karl Popper advanced the theories of conjecture and refutation where one would first observe facts, then set up theories and hypotheses to explain the observed facts, and would collect data to either refute theories till the best theory would survive. To paraphrase Sherlock Holmes, this would be akin to consider all possibilities, and then rule out the impossibles, till whatever would remain, however improbable, would be accepted as the truth. Sherlock Holmesian ideas have prevailed in modern science and have been used in evidence based health and healthcare \cite{book:650018}. Here are a couple of examples as to how to use the ideas of conjecture and refutation in Evidence based medicine. Not all research are causal in nature. In order to differntiate causal from non-causal research, we need to take into consideration the followng four entities:Any association must be a true valid association. A valid association has to fulfil three characteristics as follows.A valid association between X and Y would mean that the observed association between X and Y could not have arisen by chance alone. Therefore, in establishment of this, one has to rule out the play of chanceWe should be able to eliminate all biases in the pattern of the observation of an association between X and Y. What this would mean is, in setting up the observations that support an association between X and Y, this association cannot be explained by any systematic observation error on part of the investigator or cannot occur due to erroneous reporting or both.All alternative explanations in the form of confounding variables must be controlled for (this is the principle of controlling the confounding variables)Finally, the nature of such a causal association can be examined in the light of several considerations that Sir Austin Bradford Hill proposed in a paper he presented in 1965. Although these are referred to as Hill's criteria, according to Sir Austin Hill himself, these were considerations. We rule out the play of chance either or both at the stage of planning the study to obtain data to support or refute our hypotheses, or during the phase of analysis. This brings up the topic of hypothesis testing, p-values, and issues such as confidence intervals which we discuss presently.We should eliminate all forms of biases when we conduct studies. Biases can only be controlled at the stage of planning a study. You can examine the presence of biases following data analysis. Different forms of biases are selection bias, response bias, information bias. Biases can be differential misclassification bias or non-differential misclassification bias where the effect size is biased towards null value. In clinical studies and in evidence based health, blinding is a way to eliminate biases. Blinding refers to the situation where the investigator, the participants in the trial, and the analysts are "hidden" or deliberately made unknown or unidentifiable. Thus blinding can be single blind, double blind, or triple blind. Despite deliberate blinding efforts, it is possible to reveal the identity of the participants or know about the identity of the paticipants that compromise with the blinding efforts. In observational epidemiological studies, training of the data collectors, interviewers, use of standardised questionnaires, and objective measurements are used to minimize the risk of bias. We should control for all possible confounding variables in the association between X and Y. Confounding variables are those variables that are associated both with X and Y but they should not come in any causal pathway that connects X and Y. The different ways to control for confounding variables include stratified analysis, matching, randomisation, and multivariate analysis of data. Even if we have established that in deducing the association between two entities, we have ruled out the play of chance, we have eliminated biases, and we have controlled for confounding variables, that would not be sufficient to establish the nature of this association, whether such an association is causal or non-causal. This can be partly tested by using the considerations that Sir Austin Hill talked about in 1965 . Thesea re re are referred to as "Hill's criteria", although strictly speaking these are not criteria at all. These considerations are as follows:Strength of association .... Strength of association is a function of two parameters: prevalence of the exposure and a measure of the association. The strength of association in case of intervention studies is best provided by attributable risk (AR); this is also referred to as abolute risk. From the absolute risk reduction, we use another measure referred to as Population Attributable risk or proportions of this. The population attributable risk is given by the following formula