Health Data Science Flashcards
(26 cards)
Give some categories of health data
Patient data Specific instruments (Questionnaires, rating scales) Data from blood and tissue samples Data from images Health and fitness devices
What is measured in a cross- sectional study?
Measures variables of interest at the same time
eg classically exposures (Risk factors) and outcomes (disease)
Give some examples of cross-sectional study?
- Prevalence studies
- Aetiological studies
List some strengths of cross-sectional studies?
- Relatively easy/cheap to conduct
- Provide distribution/burden of exposure/outcome information
List some weaknesses of cross-sectional studies?
- Only measures prevalence, not incidence
- Can be difficult to establish time-sequence
What is a case-control study?
Starts with cases and controls and look to see who had the exposure (Risk factor) in the past
Often used for diseases with a long latent period.
List some strengths of case-control studies?
- Quick and relatively cheap (compared to cohort)
- Good for studying rare diseases
- Good for diseases with long latent periods
List some weaknesses of case-control studies?
- Prone to selection bias (ie unrepresentative controls)
- Prone to information bias
- Cannot establish the sequence of events
What is a cohort study?
- Aetiological research - people without a disease, risk-factors measures and then follow-up for disease
- Prognostic research - People with a disease, characteristics measured, follow-up for outcomes
List some strengths of cohort studies?
- Exposures/Risk factors measured at start of study before outcome occurs - no measurement bias.
- Can provide data on time course
- Multiple outcomes can be measured
List some weaknesses of cohort studies?
- Slow and potentially expensive
- Inefficient for rare diseases
- Exposure status may change during study
- Differential-loss to follow up may introduce bias
What would be the gold-standard interventional study?
Randomised controlled trial
What are the benefits of proper randomisation?
- Comparison groups should be similar with respect to confounders, both measured and unmeasured
- Prevents bias in the allocation of participants to treatment/control
- Only difference between groups should be if they received the intervention - therefore any difference should be attributable to the intervention
What should be considered in RCT risk of bias?
- Was randomisation sequence unbiased?
- Was allocation concealed until enrolment?
- Were participants/outcome assessors aware of treatment group?
- Have participants deviated from intended interventions?
- Are they missing data which could introduce bias?
- Was measurement of outcome unbiased?
- Was the pre-specified primary outcome reported?
What are some opportunities provided by ‘big data’?
- Wide applications - predictive modelling, clinical decision report, safety monitoring, public health
- Potentially more comprehensive data
- More-detailed data (eg wearable devices)
- Costs/efficiency
What are some challenges provided by ‘big data’?
- Privacy/security
- Quality of data (Missing data, Biases)
What is machine learning?
An automated way to find patterns in data without being explicitly programmed where to look or what to include
What is deep learning?
Subset of machine learning using more complex computational techniques to learn complex patterns in large amounts of data
What is supervised machine learning?
Training a machine by showing it examples instead if programming it.
Parameters can be altered.
List some applications of deep learning in healthcare?
-Diagnosis (Automated fracture detection, categorisation of benign vs malignant/histology/skin lesions etc)
- Data monitoring in ICU
- Prognostication
List some ethical principles relating to health data?
- Privacy
- Public interest
- Consent
- Transparency
- Security
- Proportionality
- Identifiability
What can be done to reduce the effect of chance in a trial?
Increasing sample size
What does a low p-value mean?
Lower the p-value, the less likely that a particular finding is due to chance
What is selection bias?
Systematic error in selecting study population.
ie Those recruited are not representative of reference population, or comparison groups are not comparable
