Statistical terms

These are attached with the aim of indicating areas, which you need to tryand grasp. You also need to have some idea of what a Meta-analysis is and how tointerpret it. Trisha Greenhalgh wrote a series of excellent articles on”How to read a paper” in the BMJ. These would be well worthselectively reading in preparation for this section of the examination.

Statistical terms

Prevalence This is the number of cases of a condition in the population at a particular point in time.
Incidence Numbers of new cases of that condition that occur during a defined period in a defined population.
Specificity and sensitivity Applies to diagnostic tests.
Sensitivity Proportion of true positives that are correctly identified by the test.
Specificity Proportion of true negatives that are correctly identified by the test. They are one approach to quantifying the diagnostic ability of a test.
Predictive values These allow judgments regarding the meaning of a positive or negative test result. The predictive values change as the proportion of people with the disorder in the population tested changes.
Positive predictive value Proportion of patients with positive test results who are correctly diagnosed.

Predictive values apply to one condition / test only and their validity depends on the prevalence of the abnormality in the patients being tested. As an abnormality become rarer so it becomes more certain that a negative test indicates no abnormality and less certain that a positive one indicated an abnormality. Predictive values are calculated using the specificity and sensitivity of a test and the prevalence of the condition. If the prevalence of the condition is very low, even if the specificity and sensitivity of the tests are high the positive prediction value will not be close to 1 and inevitably in screening The general population, many with positive tests will be false positives.

Negative predictive value Proportion of patient with negative test results who are correctly diagnosed.
Quantitative and qualitative data Quantitative Measurable (height/weight) and discrete (people in a house).
Qualitative More difficult to define. Usually more relevant to General Practice e.g., better/not better. Smoking/not smoking (be careful about accepting data on smoking/drinking habits based on subjective evidence alone).
Mode / mean / median / range and standard deviation Range Quantity variables arranged in order are called a distribution, and the distribution between the maximum and minimum is the range. (The distribution from the highest to lowest values).
Median The midway (median) value.
Mode The most common figures (figures with the highest relative frequency).
Mean The average figure (total value divided by the total number of observations). The arrangement of these figures in a diagrammatic form is called a frequency distribution.
Standard deviation (SD) This is an expression of dispersion of values from the mean in a “normal” distribution. Approximately 68% of values ‘will lie within 1 SD from the mean and 95% within 2 SD.
p values These are inferential statistical values derived from statistical treatment of the results. They are used to assess the degree of probability and to see if any observed effect due to a particular intervention is occurring greater than by chance i.e., is it statistically significant. The accepted p value of significance is <0.05 i.e., there is a 5% (1 in 20) or less chance that it has occurred by chance alone. A p value of < 0.01 is “highly statistically significant”.
Alpha and beta errors Beta errors These occur because the sample size is too small and a real, statistically significant difference does not emerge although there is a definite and real difference that is clinically important (The early streptokinase trials had this problem). “Absence of evidence is not evidence of absence”
Alpha errors Differences arise purely by chance but are statistically identified as real differences.
Confidence interval (CI) This is a shift in the attention of the inferential statistical treatment of the results to the effect actually observed in the study and calculates a region around it where the true effect is likely to be. The assumption is that the true effect is not very far from the observed effect and the 5% most extreme possibilities of change imbalance are excluded to give the 95% confidence interval. It is felt to be more useful clinically than p values.

E.g. the results show an increased relative risk of 3.5 in the non intervention group with 95% CI of 1.4-8.5. The narrower the confidence interval about the observed effect the more reliable and useful it is.

It tends to be more useful in interventional than observational studies. The 95% CI is affected by sample size and the number of events generated in the study period. With few events (a too small sample size), the C I at 95% will become so wide as to be unhelpful (despite an observed clinical effect).

Bias Faults in study design lead to a misinterpretation of the relation between exposure and disease. Some examples of bias applicable to a randomised controlled trial are:

  • Selection bias: systematic differences in the comparison groups attributable to incomplete randomisation.
  • Performance bias: systematic differences in the care provided, apart from the intervention being evaluated.
  • Exclusion bias: systematic differences in withdrawals from the trial.
  • Detection bias: systematic differences in outcome assessment.
Confounding Exposure seems to be associated with disease. However, the relation exists only because the exposure is associated with other risk factors for the disease and not because the exposure causes the disease.
Observational study In these, the natural course of events is observed.

  • Case controlled study 
    People with a particular disease are matched to a similar group without the disease in order to try to prove/disprove an associated factor. These studies identify affected and non affected individuals and then lock back at their exposure to a factor. They have a risk of Bias as disease and exposure has already occurred. They are generally retrospective.
  • Cohort study 
    Subjects are classified according to their exposure I non-exposure to a suspected risk factor and followed up over a period (usually years) At the time of exposure all subjects must be free of disease.
  • Interventional study 
    In these people who already have a risk factor are followed and matched to a treatment or non treatment group.

Critical appraisal (Nottingham VTS) (2011)


en English