6. Tutorial for POPLHLTH 216: Quantitative methods for Health

• Exam Revision
  • Measures of disease frequency
  • Probability
  • Statistical tests
  • Plotting
  • Bias
  • Sampling
  • Regression
• Short answer - Responses
• Homework

Photo by Luis Villasmil on Unsplash

Exam Revision

This session we will do a bit of exam preparation, looking over some selected questions that cover parts of the course.

The examination will be conducted online, using Inspera.

You will have 20 multi-choice questions and 3 x 10 mark short answer questions.

Measures of disease frequency

Some questions will be multi-choice. The best measure of burden of disease in a population is generally given by: Hospital admission rateIncidence ratePrevalencePopulation attributable riskCumulative incidence

Think about the definition of each of these terms. Which is best at describing the health status of a population?

For example, if you wished to know the burden of diabetes in an adult population, which measure would be most helpful?

Remember, my epidemiological glossary may help you with revision.

Probability

A doctor estimates that the pre-test probability of a patient with pleuritic chest pain in hospital having a pulmonary embolus (blood clot in the lung) of 8%. They then carry out a d-dimer test which is 90% sensitive and 30% specific, which turns out to be positive. What is the post-test probability that the patient has a blood clot in the lung?:

10%50%Not possible to estimate100%20%5%

Statistical tests

You are considering the association between a diagnosis of lung cancer and smoking status (both binary variables) in a case-control study.

An appropriate statistical test would be:

A t-testANOVAA chi-square testLinear regressionPopulation attributable risk

You are considering testing for the effect of three different dietary regimens on change in body mass index in a large randomised controlled trial. What is the most appropriate statistical test to use?

Population attributable riskLinear regressionA chi-square testANOVAA t-test

Continuous outcome and three different groups? What is the most appropriate test?

When testing for the association between bed sharing and cot death, an appropriate statistical test would be:

ANOVAPopulation attributable riskA t-testA chi-square testLinear regression

The condition associated with a P-value is:

That confounding is not presentThat confounding is presentThe null hypothesis is trueThe exposure causes the diseaseThe null hypothesis is false

Assume that in a cohort study there is a strong association between the presence of rotten teeth and cardiovascular disease. Once the population is then subsetted by high and low sugar intake, there is then no association between rotten teeth and heart disease in each of the sugar exposure groups. Which is the best explanation of this phenomenon?

Effect modificationRandom errorSelection biasConfoundingMeasurement error

Plotting

What is the best plot for assessing the relationship between two continuous variables?

HistogramBarplotBoxplotDotplotScatterplot with smoother

Bias

A dermatologist tells you that they frequently miss the diagnosis of scabies in patients who eventually turn out to have the disease. Which of the following epidemiological statements are match this clinical observation?

None of theseThe sensitivity of the dermatologist's exam is lowThe sensitivity of the dermatologist's exam is highThe specificity of the dermatologist's exam is highThe specificity of the dermatologist's exam is low

Which of the following methods is most likely to reduce information bias in a cohort study?

Increasing statistical powerAdjusting for confounding during the analysisChecking for effect modification with other baseline exposuresWriting a clear protocol of how methods of measurement are to be conducted during the studyUsing regression modelling rather than stratification methods

Sampling

In a designed survey of diabetes prevalence, which of the following methods is most likely to increase the precision of the estimate for a given sample size?

Snowball samplingCluster samplingStratified samplingConvenience samplingSimple random sampling

Regression

In a linear regression which investigates the relationship between age and blood pressure, which of the following parameters will best describe the measure of association between the two variables?

The R^2 of the linear model in which age is the independent variable and blood pressure the dependent variable.The y-intercept in which age is the independent variable and blood pressure the dependent variable.The y-intercept in which blood pressure is the independent variable and age the dependent variable.The beta coefficient of the linear model, in which blood pressure is the independent variable and age the dependent variable.The beta coefficient of the linear model in which age is the independent variable and blood pressure the dependent variable.

Short answer

1. In 2007 a case-control study was undertaken to determine whether alcohol consumption in the 6 hours before injury was a significant contributor to fall-related injury. Cases were derived from hospital admissions identified by study staff as having suffered a fall-related injury. Controls were randomly derived from the electoral roll in the community which served the hospital, with a similar age distribution to cases. Exposure period for controls was randomly allocated to a 6 hour period in the last 24 hours. Alcohol exposure information was derived from self-report. The study results are printed in the summary table below:

Alcohol intake	Cases	Control
Yes	40	70
No	18	87

1. With the following set of formulae, derive the odds ratio and 95% confidence interval. Also estimate a population attributable risk.

(Show your working).

\[\begin{align} \text{Standard error}( ln(\text{OR}) ) &= \sqrt{1/a+1/b+1/c+1/d} \\ \text{Error factor} &= e^{[1.96 \text{ x } \text{s.e.}(ln(\text{OR})]}\\ 95 \% \text{ CI} &= \text{OR} / \text{EF to OR x EF}\\ \end{align}\]

And population attributable risk is given by the following formula:

\[ \text{Population attributable risk} = \frac{P(E) (\text{RR} - 1)}{1 + P(E) (\text{RR} - 1)} \] Where:

• \(P(E)\) is the prevalence of the exposure in the population, and
• \(\text{RR}\) is \(\frac{P(D \mid E)}{P(D \mid \bar{E})}\)

Responses

Odds ratio =

Standard error( \(ln\)(OR) ) =

Error factor =

Lower CI =

Upper CI =

Warning
Since this is a case-control study we will estimate prevalence of exposure from controls. In the case of a cross-sectional or cohort study it is the unconditional or overall prevalence of exposure.

Prevalence of exposure in controls (as a proportion) =

PAR (proportion) =

(4 marks)

2. Describe the meaning of the odds ratio and 95% confidence interval in words to a news reporter interested in the study. (2 marks)

3. A representative from the alcohol industry criticises the study because the exposure is measured by self-report. Comment on the likely effect of such a bias, and how it might be investigated further. (4 marks)

2. Explain the meaning of a P-value, when used to interpret the effect measure from an observational study. (2 marks)

2. Which factors influence the size of the P-value? (2 marks)

3. When associated with an effect estimate, discuss the meaning of a 95% confidence interval. (2 marks)

4. Compare and contrast the use of P-values and 95% confidence intervals, when reporting epidemiological study results. (4 marks)

3. (a) You are reviewing a range of epidemiological evidence which claims a positive association between exposure to passive smoking and the incidence of lung cancer for the Ministry of Health. This organisation is considering recommending that the government pass progressive Smokefree legislation. You have in front of you a submission from the tobacco industry that argues that the effect estimates for this exposure-outcome association is weak and that since the studies are mainly observational, it cannot be proven that passive smoking is linked to lung cancer. Your employer wants you to contradict this argument. Outline your response, including theoretical frameworks which may help formulate your answer. (7 marks)

2. You have estimates of the prevalence of exposure to passive smoking and an estimate of the effect of passive smoking exposure on lung cancer incidence. Outline an epidemiological measure that may help extend the case for further restriction of passive smoking. What assumption underlies such a calculation? (3 marks)

Homework

Check out some old examination questions here. My exams cover those from 2020 to present day.