Skip to content
Home » Peri-op & Misc » Statistics

Statistics

Diagnostic tests

 

Sensitivity

Ability to detect disease, true positives  (if the patient has the disease, how likely is the test able to detect it)

True positives / true positives + false negatives

i.e., how many people who actually have the disease, did the test pick up

 

Specificity

Ability to rule out disease or confirm normality, true negatives

True negatives / true negatives + false positives

i.e., how many people without the disease did the test correctly identify as normal

 

Positive predictive value

If the test is positive, how likely is it true the patient has the disease

True positives / true positives + false positives

i.e., is the test is positive, how likely is it to be true

 

Negative predictive value

If the test is negative, how likely is it true the patient is disease-free

True negatives / true negatives + false negatives

 

 

Risk

Relative risk

Occurrence (risk) in exposed group / occurrence (risk) in comparison group

 

Absolute risk difference

Occurrence (risk) in exposed group – occurrence (risk) in comparison group

 

 

Number needed to treat

  • Number of subjects who must be treated with exposure or intervention, to have one subject experience the outcome
  • 1 / risk difference ARR (in absolute %)
  • i.e., if absolute risk reduction is 2 %, NNT is 1 / 0.02 = 50
  • if absolute risk reduction is 20 %, NNT is 1 / 0.2 = 5

 

 

Odds ratios

Odds ≠ risk or probability

Odds = probability of something happening / something not happening

Odds of 1:4 imply a 20% risk of something

Odds of 1:1 imply a 50 % risk of something

 

Odds ratio = ratio of the odds between two groups.

Not the same as relative risk due to above.

 

 

Incidence and prevalence

Incidence

Number of individuals who develop a disease or condition within a specified period of time

Incidence only includes new cases developing during the time analysed, and still identifies cases which die early or are cured

Prevalence

Proportion of individuals who have a disease or condition at a specified point in time

Prevalence includes all cases – pre-existing or newly diagnosed, but may miss cases who die rapidly or are cured before specific timeframes analysed

 

 

Accuracy and precision

 

Outcome measures

Intention to treat analysis

Includes all patients in analysis based on initial allocation (regardless of crossover, dropout)

More closely aligns with real world practice

Does not undermine randomisation process

 

Per protocol analysis

Includes only those patients who underwent the planned intervention correctly

May give over-optimistic results regarding the intervention

 

T tests

Differences in means / standard error of the differences in means

Standard error calculated from SD and number of subjects

Population must be normally distributed, continuous variable and same populations

P value halves for one sided test cf. two sided (two sided would be preferred)

 

Non parametric tests

For non ordinal / categorical / non continuous outcomes or non-normal populations

Chi-square test or Fisher’s exact test

 

Outcomes

Should be clearly defined, measurable, relevant and reliable

Surrogate endpoints should be interpreted with caution

 

Internal validity

Extent that a measure or test is correct for the group of patients evaluated

External validity

Ability of a measure or test to be applied in routine clinical practice – generalisability

 

 

Critical appraisal and bias

 

Bias – features of study design, conduct or analysis which result in estimates of effect which may be different to the true population effect

 

Selection bias

  • Subjects studied are not representative of the eligible population
  • e.g. volunteers likely to be better educated, surgical populations self-selected as fit and healthy, patients unlikely to be compliant are excluded, non-English speakers excluded,

 

Confounding bias

  • A variable associated with the exposure factor influences the outcome
  • e.g. surgeon not controlled for in surgical trials, older patients or those with advanced disease appear to do worse etc.
  • Can be minimised with randomisation or stratification analysis

 

Measurement bias

  • May result from errors in measurement of patient factors, or disease outcomes
  • Examples include measurement of factors for selection criteria, patient compliance or exposure to treatment, lost to follow up, blinding of measurement of subjective outcomes