Clinical trials

in a clinical trial, what is the independent and dependent variable

independent

• intervention → what we do to the patient e.g. give drug and compare with placebo

• exposure → what happens to the patient that we don’t intervene with e.g. did pollution affect the outcome/endpoint

dependent

• outcome or endpoint e.g. compare the quality of life of the treated vs untreated

causality → does the intervention/exposure cause the outcome endpoint?

what are the different classes of clinical studies

• observational → we observe what happens naturally and don’t intervene

• experimental → we intervene and give some type of intervention (the most useful experimental study is the randomised controlled trial, RCT)

how are clinical studies graded

using the hierarchy of evidence 

• case reports, anecdote, personal opinion → not a good way to make decisions on clinical practice 

• meta-analysis → combines the results from a number of studies into one, than a single study alone 

what techniques are used to make guidelines, considering the strength of the evidence

• class (strength) of recommendation

• level (quality) of evidence

class (strength) of recommendation

level (quality) of evidence

what are types of observational studies

case-controlled study

• exposure is unknown → compare people with a disease to those without it and → start the study 

• e.g discovering the links for lung cancer and if they have a history of smoking

prospective cohort

• exposure is known → compare the exposed vs unexposed groups over time → start the study before the disease occurs → see who develops the disease

• e.g. smokers vs non-smokers and see how many have lung cancer 

retrospective cohort 

• exposure and disease are known and have occurred → look backwards at old records to show how exposure over time is related to the development of a disease 

• e.g. look at records to compare heart disease rates between patients who were given Drug A vs. Drug B, who developed the disease?

examples prospective cohort study: does owning a dog reduce your risk of heart disease

• take a group of people and find out which of them are dog owners and not dog owners 

• follow up over ten years to see who develops heart disease 

• results may show that people that aren’t dog owners have a higher incidence of heart disease 

• we can’t conclude that owning a dog reduces risk of heart disease as you have to make the assumption that both groups are identical in everything except whether they own a dog or not 

• look at other characteristics in the individuals e.g. how much walking they do: from sedentary → extensive walkers 

• however, people with dogs walk more, which explains the relationship → owning a dog doesn’t reduce the risk of heart disease but going for a walk 

• this is known as confounding

how are observational studies prone to confounding

cofounders are variables related to both the exposure and the outcome

• in the example study before, walking confounds the relationship between dog ownership and disease 

• if we know about and measure confounders, we can apply statistics to correct for them

• however, observational studies always have some ‘residual confounding’ → observational studies determine association, not causality 

what are randomised controlled trials, RCT

example: which drug is best at preventing heart disease

• test drug on a sample of people at risk of heart disease in a population

• randomly allocate people to a certain drug treatment

• follow up overtime and measure outcomes

• there are now two identical populations, except in respect of the experimental intervention → randomisation eliminates confounding → RCTs demonstrate causality 

what is a systematic review 

• a literature review that focuses on a research question that tries to identify, appraise, select and synthesise all high quality research evidence relevant to the question 

• systematic reviews that have high-quality RCTs are crucial to evidence-based medicine 

what is meta analysis

combining the results of several independent studies for the purpose of integrating the findings

• results come from multiple underpowered clinical trials 

• it uses existing data to test new hypotheses, which may not have been considered by original investigators 

what are 95% confidence intervals

• 19/20 of the confidence intervals will be correct → true population mean is enclosed by the interval

• the remaining one will be wrong → an unrepresentative sample 

why are statistics useful

• can describe quantities (Descriptive statistics and effect size e.g. mode, median)

• can decide whether there is a difference between two (or more)
  groups: hypothesis testing

• if we are evaluating an antihypertensive drug, we need to know:

  • whether differences between the groups are due to the drug, or due to chance (hypothesis testing).

  • how much blood pressure is reduced by the new drug (effect size)

null hypothesis vs alternative hypothesis

• null → no difference exists between data sets 

• alternative → opposite of the null hypothesis 

scenario

• we conduct a trial to compare the antihypertensive effects of Drug A and Drug B

• hypertensive patients are randomised to receive Drug A or Drug B

• at the end of the trial, systolic blood pressure is 120 mmHg in patients treated with Drug A, and 125 mmHg in those treated with Drug B

• is Drug A more effective than Drug B?

• is the difference (5mmHg) a real effect, or can it be explained by
  random variation (chance)

• null Hypothesis: No difference between effectiveness of A and B

• alternative Hypothesis: there is a difference

• we need an objective way to decide which hypothesis to accept. We are at risk of:

  • Type I ( α) error (Rejecting the null when it is true – ‘False Positive’)

  • Type II (ß) error (Accepting the null when it is false – ‘False Negative)

what is p-value

• the probability of obtaining results as extreme as (or more extreme than) the observed results assuming that the null hypothesis is correct.

• a smaller p-value → less likely that results are due to chance → stronger evidence in favour of the alternative hypothesis 

what is p<0.05

many researchers use p=0.05 as a threshold for ‘statistical significance’

• they reject the null hypothesis if P<0.05 → we therefore accept a risk of 5% * of rejecting the null when it is true (Type 1 ( α) error )

• Many people think P<0.05 is too lenient

• However, P<0.05 is equivalent to non-overlapping 95% CIs

(relates to ideal experimental conditions, in most situations the risk is >>5)

when p<0.05, what else do you need to consider

• where P < 0.05 all you have produced is evidence that an effect does exist → should always consider the size of the effect

• with a measured end point – How much does the mean value change as a result of the change in treatment?

• with a classified end point – How great is the change in the proportion of individuals falling into each category?

what determines how to quantify data

variable type e.g. effect of drug A and B on BP and mortality

what tests do you use for different types of data