Semester 2 - Lecture Revision Flashcards

Question

What is likely the greatest problem/fault with calculating disease burden?

Answer 1

Perhaps the greatest weakness is the reliability of the underlying data on which estimates depend.

Answer 2

We take a sample of people or patients. We conduct the index test, which is our new test that we are assessing. And then we conduct the reference test, which is best way to identify if people have or do not have a disease What we end up with for each individual in our study population, is a result for the reference test and a result for the index test

Answer 3

Data from the study shown in a 2x2 table We can see that 11 of the 13 participants who were HIV+ on the Western Blot, tested HIV positive using the OraQuick test - true positives 2 out of the 13 participants who were HIV+ on the Western Blot were incorrectly identified as HIV- - false negatives. Of the 1061 participants that were HIV- according to our reference test, 1060 of these were negative using the OraQuick test - are our true negatives. 1 person who was HIV- but tested HIV+ using the OraQuick test - false positive.

Answer 4

The closer the sensitivity is to 100%, the better the test is at ruling out disease when the test is negative - is the test good at picking up treu positive results.

Answer 5

The closer the specificity is to 100%, the better the test is at ruling in a disease when it is positive - is it good at picking up negative results.

Answer 6

Out of the positive results give by the new test, how many are true positives. It is important to note that positive predictive values are affected by the sensitivity and specificity of the result, as well as the prevalence of disease in the population. The positive predictive value will increase with increased prevalence of disease

Answer 7

Like the positive predictive value, the negative predictive value is affected by the sensitivity and specificity of the result, as well as the prevalence of disease in the population. The negative predictive value decreases with increased prevalence of disease.

Answer 8

**Likelihood ratio for positive results**, and is calculated by taking the ratio of the sensitivity for a test to 100 minus the specificity **If this ratio is greater than one, then the test result is associated with disease** – and we can see for the OraQuick test that having a positive test result is associated with disease with a likelihood ratio for positive results of 846. Essentially, people with HIV are 846 times more likely to have a positive test than someone without HIV. **Likelihood ratio for negative results** - This is calculated by taking the ratio of 100 minus the sensitivity to the specificity for a test. **If this ratio is less than one, then the test result is associated with not having disease** – and we can see for the OraQuick test that having a negative test result is associated with not having disease with a likelihood ratio for 0.15.

Answer 9

Representation of diagnostic test performance Plot sensitivity vs. 1-specificity Perfect test - area under the curve = 1 Test as good as chance - AUC = 0.5

Answer 10

We measure the analyte of interest in a sufficiently large sample of **healthy individuals**. This is our reference population, and it should take into account factors such as age, sex, and ethnicity. We plot the values obtained against their frequency of occurrence. For analytes that occur in a Gaussian distribution, the normal range is defined as 2-standard deviations - ecompasses 95% of the values in the reference population

Answer 11

Same applies - We use the upper and lower 2.5 percentiles as the cut-off points for our reference range.

Answer 12

**Sensitivity** is how good the test is picking up an individual with the disease. **Specificity** measures how good is the test at identifying the absence of disease.

Answer 13

The other aim is to investigate the effect of potential risk factors on the frequency of disease We want to assess whether a disease is more common in one exposed population compared with another unexposed population.

Answer 14

First two columns there are only two response options – this is known as **binary data**, and is commonly what we work with in epidemiological analysis. Third column - **categorical** - more than two response options - if the categories can be listed in order, as we can here, then this is known as an ordered categorical data. If the categories may be listed in any order, then this is called unordered categorical Fourth column - **numerical or quantitative data** -numerical data can be classified as either discrete (whole numbers) or continuous.

Answer 15

Association can be used to describe the relationship between two variables, in which a change in one variable is associated with change in the other. Look at the association between exposure and outcome How we measure association depends on the type of data we have for our exposure and outcome

Answer 16

Measures of effect compare the frequency of disease in exposed and unexposed groups. **Risk ratio** - risk of disease in the exposed group divided by the risk of disease in the unexposed group **Risk difference** – also known as the attributable risk - is the excess risk in the exposed group that we estimate is due to being exposed – so to calculate this we take away the risk in the unexposed group from that in the exposed group **Attributable risk** - calculated as the attributable risk, divided by the risk of disease among the exposed - interpreted as the % of the outcome which is attributable to the exposure amongst the exposed group.

Answer 17

What is the impact of... 1. Chance 2. Bias 3. Confounding variables

Answer 18

Used to make sure that the associaiton seen does not arise by chance

Answer 19

95% confidence interval Larger our sample size the smaller the confidence interval as we can be more certain in our observed estimate.

Answer 20

Different types of bias that can be introduced as part of the study design and can lead to an incorrect estimate of the effect of an exposure on an outcome. **Selection bias** refers to error due to systematic differences in characteristics between those who take part in a study and those who do not, such that either the **comparison groups are not comparable** or the people recruited into the study are somehow **different from the population** that should have been included in the study. **Information bias** is any error in the measurement of exposure or outcome that results in systematic differences in the accuracy of information - measurement error

Answer 21

A third factor that is independently associated with both our exposure of interest and our outcome, but is not on the causal pathway.

Answer 22

A **correlation coefficients** (also called r in statistical notation) can be estimated to describe the association between two continuous variables such as weight and blood pressure. When the outcome of interest is a categorical variable the distribution of exposures/risk factors can be compared in people with and without the outcome using descriptive statistics such as **t tests or chi squared tests**

Answer 23

It is important to be aware that there is not a statistical test for causality and that all the relevant evidence needs to be considered to review the strength of support for a causal relationship Causality is easeir to establish for communicable disease - Henle Koch's Postulate Harder to establish for non-communicable disease

Answer 24

Define prognosis as an opinion based on medical experience of the likely course of a medical condition

Answer 25

Most commonly we measure prognosis in terms of survival, whether that be disease-free or overall survival.

Answer 26

Absolute rates are probably the most commonly reported outcome measure in prognostic studies. This is percentage of patients with a given outcome at a particular point in time. Examples 1. Median - time point at which 50% of patients have experienced the outcome 2. Five-year survival - % of people surviving after 5 years Simple but does not convey information well

Answer 27

Relative rates - allows us to determine the risk of a particular individual developing an outcome of a disease compared to another individual based on particular characteristics. Reported as an odds ratio or a hazard ratio from a multivariable model. Example Patients undergoing surgery in low-income groups were more likely to develop a surgical site infection than those in high-income settings - odds ratio of 1.6 or 60% increased risk

Answer 28

Survival curves - allows us to graphically determine the cumulative events over time. The left graph, we have a y-axis which is some outcome, which here is cumulative survival. Where one means all patients are alive and zero means that all patients unfortunately have died. And on the x-axis we have time. This allows us for each time interval to determine the probability of patients surviving a particular disease. Right graph - effective way of comparing two different treatments

Answer 29

New diagnosis tool made - allows for earlier diagnosis which would appear to increase prognosis but in reality this is not true - just diagnosing earlier

Answer 30

Usually they're a collection of variables which are combined together and modelled in a particular population to predict an outcome or disease. So for example, we may have a prediction model which contains five variables - used to predict disease severity on admission to hospital Used for... 1. Aid in medical decision making 2. Supplement clinical investigations and/or additional investigations 3. Help stratify patients into groups 4. Provide thresholds for drugs and treatment

Answer 31

AUROC - sometimes known as concordance, C-statistic or discrimination - is the most commonly used metric to measure performance. This is quite a classical graph demonstrating that AUROC - allows for comparison of risk tools X-axis - false positive rate Y-axis - true positive rate Dashed line here, which shows in AUROC - or AUC - value of 0.5 - if a score runs across this dotted line, then it's no better than chance at predicting the outcome in this set of patients

Answer 32

The reason calibration is important is that it estimates and demonstrates the risk profile across the entire cohorts. Example If we see on the left-hand side of this curve, for patients with a predicted probability of 0.0 to 0.1, they're calibrated quite well (i.e. they run across this dotted line, which means that the predicted versus observed risk is about the same). But as we move up in terms of predicted and observed probability of the outcome, we can see that it's skewed to the right. Now, this is important because what we need to know for a tool is does it **over or underestimate the risk of the outcome**?

Answer 33

In clinical trials we undertake a scientific experiment where we intervene to allocate our study population to be exposed or unexposed and, if this is done well, this should mean that the exposed and unexposed groups are comparable across a range of characteristics. As noted here clinical trials are often referred to as intervention studies or randomised control trials.

Answer 34

PICO stands for patient or population, intervention, comparison and outcome.

Answer 35

Selection bias occurs if the study population does not reflect a representative sample of the target population

Answer 36

Type of intervention - doesn't need to be a new medication Specifics of intervention - dose, frequency, etc. It is also important to consider whether the trial is aiming to measure efficacy or effectiveness of an intervention **Efficacy** - Efficacy can be defined as the performance of an intervention under ideal and controlled circumstances, such as a clinical trial. **Effectiveness** - effectiveness refers to performance under real-world conditions or in average clinical conditions - referred to as a pragmatic trial

Answer 37

Big thing - Randomisation! Allocation to treatment or control arms - ensures fair comaprison and avoids selection bias Many potential methods

Answer 38

The choice of a control intervention is a critical factor in the design, conduct and interpretation of the clinical trial. A comparator could be... 1. No intervention 2. It could be a placebo or sham treatment which is an inactive therapy made to look identical to the active therapy. 3. Or the comparator could be another active intervention or the usual treatment Three stypes of comaprison to active intervention - Superiority - superior of other - Equivalence - not too different - Non-inferiority - not much worse Note - trials are usually double blind - the trial participant and investigator/evaluator do not know which treatment a participant is receiving

Answer 39

When reading or designing a clinical trial we need to be clear on what the primary and secondary outcomes are to answer the trial objectives. These are outcomes that are measured during the course of the trial, and they define and answer the trial questions. In general, a single outcome should be used to answer the primary objective, but there may be a number of additional outcomes (including on potential side effects of the intervention) and these are generally referred to as secondary outcomes.

Answer 40

Specific measure that is often calculated for trials of preventive interventions - the protective efficacy We can then calculate the protective efficacy (or effectiveness), by subtracting the risk in the intervention group from the risk in the placebo group, and then dividing this by the risk in the placebo group Example - paracetamol reduced the number of headaches by 60%.

Answer 41

More expenditure **does not** necessarily equal better healthcare. Plus there it's a law of diminishing returns. So what is absolutely clear is that we need a method for efficiently allocating our resources, if we want to maximise health gains for the population from our health care budget

Answer 42

Key feature of opportunity cost in healthcare is the opportunity cost is not readily identifiable against individuals. As clinicians, we're not forced to choose between individuals. We simply make a decision about whether or not allocate resource

Answer 43

Macro level or aggregate level - This is thinking about how much healthcare should we provide for the population and how do we organise care at a system level to maximise the health care At the micro or individual level, this is thinking about individual patients, or indeed an average patient out of a population - we're generally making a trade-off between different treatment options and ultimately we’re deciding who should gain from investment decisions and who shouldn't, who should live and die.

Answer 44

So in health economics, the key focus of what we're trying to achieve is measuring opportunity cost (i.e. valuing health care). The key tool of the trade that we use for this is cost-effectiveness analysis. We think about cost-effectiveness with respect to a threshold, which is otherwise known as a **willingness to pay threshold**. Above this threshold, we should be willing to pay for technology. But if something is less cost-effective than a threshold, then it's considered of insufficient value or incurs too much opportunity cost to be worthwhile in the NHS

Answer 45

Over time our cost-effectiveness threshold rises as we adopt new, more cost-effective technologies. We can see that in order for a new technology to be considered cost-effective, it has to, as a minimum, be more cost-effective than our current least cost-effective technology.

Answer 46

The cost-effectiveness threshold is the maximum amount the health service will pay per unit of health gained

Answer 47

We can combine length of life and quality of life into a quality adjusted life year or QALY, which is simply an... expected lifespan, multiplied by the quality of life weight averaged over that lifespan

Answer 48

Incremental cost effectiveness ratio, or the ICER, which compares the cost effectiveness or cost per QALY of a new intervention compared with current standard The ICER equals **new cost minus current standard care costs** divided by **new health benefits minus current standard health benefits.**