Data Flashcards
Audit cycle - 6 steps
- identify the issues
- Obtain / define standards
- Collect data
- Compare performance with standards
- Implement change
- Re-audit
Six data protection principles (GDPR )
Be processed lawfully, fairly and in a transparent manner
Be processed for specified, explicit and legitimate purposes (and nil outside this)
Be adequate, relevant and limited to what is necessary in relation to the purposes
Be accurate and up to date
Not be kept for longer than is necessary
Be secure
GDPR Article 6
FOR PERSONAL DATA
Subjects must have consented to use of their data (Consent is not recommended for use in the health sector as consent cannot be considered freely given if access to health and social care depends on it: use of Common Law Duty of Confidentiality instead. )
OR processing of data must be necessary in for one of the following:
1. For contract
2. For legal obligation
3. For vital interests
4. For task in public interest or official authority
5. For legitimate interests
GDPR Article 9
FOR HEALTH DATA “special data category” (need one from category 6 plus one of the following):
The processing is NECESSARY FOR MEDICAL PURPOSES where the processing is undertaken by a health professional or someone else who owes an equivalent duty of confidentiality.
Information on a patient’s health record is likely to be special category data (Article 9)*
Common Law Duty of Confidentiality (CLDC)
Used by the health service to store and share patient information, with and without patient consent.
Consent under the CLDC falls into 2 categories:
Implied consent – the case for most healthcare services where patients must assume their data is being used to support their care and treatment ie discussion at MDT meetings, referral to other clinicians/ specialties
Explicit consent – where the patient has agreed for the use of their data for an additional specific purpose after they have been fully informed ie research or teaching
Breaking confidentiality and sharing information under the CLDC must meet one of the following conditions:
(1) Explicit or implied consent to do so (most cases)
(2) Mandatory legal requirement / power that enables the CLDC to be set aside
(a) Safeguarding concerns (Children’s Act 1989)
(b) Notifiable illnesses and reporting of food poisoning.
(c) Care Quality Commission inspections
(d) Sharing to Health and Social Care Information Centre (HSCIC): Under the powers given to NHS Digital through section 259 of the Health and Social Care Act 2012
(3) An overriding public interest to share : Benefits of sharing the information deemed to outweigh the right to privacy of the patient and the possibility of damage to trust in the profession by breaking confidentiality
(4) A court order for the sharing of specific information and to whom
(5) The Confidentiality Advisory Group (CAG) has given Section 251 approval for the use of confidential information by the Health Research Authority (HRA) or Secretary of State for Health and Social Care
Legal support for the use of confidential patient information without consent is given
(1) Under the Health Services (Control of Patient Information) Regulations 2002
(2) Within section 251 of the NHS Act 2006
- Protects the interests of patients/ the public whilst also making sure relevant information can be used when it is appropriate for reasons beyond individual care
- Usually only granted when it would be very difficult or impractical to seek the consent of every individual whose data they wish to use: National data opt-out is offered to members of the public; Only applies to data being shared under Section 251
Patient request for medical records
Subject Access Requests (SARs) are made under the Data Protection Act 1998
The SAR does not need to be in writing, it can be verbally or electronically
You must be provide the information within 28 days. Unless exceptional – 2 month extension can be granted. The patient must be informed of this extension prior to the initial 28 days.
It is a criminal offense to amend or delete records in response to a SAR.
Exemptions and information that can be redacted:
(1) Anything that you believe may cause serious harm to the patient
(2) Any third party information
(3) Information relating to the storage of gametes / embryos (Human Fertilisation and Embryology Act 1990 UK – Section 33A)
(4) Information relating to an individual being born as a result of IVF (Human Fertilisation and Embryology Act 1990 UK – Section 33A)
(5) Where disclosure is prohibited by law ie adoption records
Legal parent’s have access to children’s records providing this is not contrary to the child’s best interests or a competent child’s wishes
Children and young people with capacity have the right to request access to their own records and also to block access to their records by parents
- In England anyone over the age of 16 is legally presumed to have capacity, and for children younger than 16 capacity should be assessed on a case by case basis
- In Scotland anyone over the age of 12 is legally presumed to have capacity
‘Next of kin’ have no right to request record access or consent to information sharing on the patients behalf (unless legally in place ie Advanced Decision/LPA)
A patient with capacity can authorise a solicitor to request access to their records, but in this instance the patient’s written consent must be gained before release
Gender Recognition Act – Section 22
protects information relating to a person’s gender history after they have legally changed gender
NHS (Venereal Diseases) Regulations 1974 & NHS and PCT (Sexually Transmitted Diseases) Direction 2000:
protect patient identifiable information relating to examination or diagnosis of STIs including HIV
Unless to another medical practitioner for the purposes of treatment
OR to prevent the spread of disease
Types of bias
Selection bias:
Attrition bias:
Measurement bias:
Observer bias:
Procedure bias:
Central tendency bias:
Misclassification bias:
Selection bias: error in the process of selecting participants for the study and assigning them to particular arms of the study.
Attrition bias: when those patients who are lost to follow-up differ in a systematic way to those who did return for assessment or clinic.
Measurement bias: when information is recorded in a distorted manner (e.g. an inaccurate measurement tool).
Observer bias: when variables are reported differently between assessors.
Procedure bias: subjects in different arms of the study are treated differently (other than the exposure or intervention).
Central tendency bias: observed when a Likert scale is used with few options, and responses show a trend towards the centre of the scale.
Misclassification bias: occurs when a variable is classified incorrectly.
P value
The probability of obtaining results as extreme as or more extreme as those observed in a study, assuming the nul hypothesis is true.
If the p value is small then you can reject the nul hypothesis and assume that the difference observed in your sample is statistically significant.
Alpha value
the cut off value that you compare your p value to, determining whether the p value is small enough to conclude that your result is unlikely to have occured by chance.
Probability of making a T1 error.
P<a = statistically significant. usually 5%.
How to calc p value base on type of variable
T test: numeric variable + categorical variable (parametric) COMPARES 2 MEANS.
ANOVA: numeric variable + categorical variable with 3 or more categories. COMPARES >2 MEANS. (anotha mean)
Chi: 2 x categorical variables (non parametric)
Correlation: 2 x numeric variables
Spearman and Pearson are correlation tests.
Pearson is Parametric. Spearman is non-parametric
Parametric vs non-parametric tests
Parametric tests assume a normal distribution of population data for the variable being tested and are used for testing variables within a population that are interval or ratio e.g. height, temperature or age.
PARAMETRIC REQUIRES the population to be of normal distribution* T TEST, ANOVA, AND CORRELATION (PEARSON)
Non-Parametric tests can be used for data that is not normally distributed within a population or is of nominal or ordinal value e.g. eye colour or marital status
CHI AND CORRELATION(SPEARMAN)
Type 1 errors
False positive
True nul which you reject
Claiming significant when not
Alpha
Increased by bias
Types of bias that can lead to a Type 1 error include: selection bias, confirmation bias, attrition bias, and nonresponse bias;
(essentially, any bias that leads to skewed data selection or interpretation, causing a researcher to incorrectly reject a null hypothesis when it is actually true, resulting in a “false positive” outcome)
Type 2 errors
False negatives
False nul which you accept
Beta
Key factors contributing to Type 2:
Small sample size:
Low statistical power:
Large standard deviation:
Effect size too small:
Strict significance level (alpha)
How to mitigate a Type 2 error:
Increase sample size:
Increase the effect size:
Use a less stringent significance level (alpha): While this increases the risk of a Type 1 error, it can decrease the risk of a Type 2 error.
Perform power analysis: Calculate the required sample size to achieve adequate power for your study.
Power
Probability of finding a difference between groups if one truely exists
1-B
Power calculations before study/ used in design
Increases with increase in
Sample size
Effect size (difference between groups)
Precision of results (decreasedSD)
Cross Sectional Study
Involves analysis of data for a population at one specific point in time
Descriptive studies
Most appropriate method to assess diagnostic tests
Likelihood ratios (LR)
Likelihood ratios (LR) are an alternative to positive and negative predictive values for estimating the likelihood of disease after diagnostic testing.
A positive likelihood ratio (LR+) is the probability that someone with a disease will have a positive test result divided by the probability that someone without a disease will have a positive result.
LR+ = sensitivity/ (1-specificity)
A negative likelihood ratio (LR-) is the probability that someone with a disease will have a negative test result divided by the probability that someone without a disease will have a negative result. The formula for a positive likelihood ratio is:
LR- = (1-sensitivity) / specificity
As a general rule, an LR+ greater than 10 and an LR- less than 0.1 show that a test reliably discriminates between people who do and do not have disease. If we know the pre-test probability of disease, we can use LR+ to calculate the post-test probability of disease among those who test positive using the following formula: Post-test odds = pre-test odds x LR+.
Radiopedia
Standard Error
The Standard Error (SE) measures the amount of variability in the sample mean OR another way to put it is the SE is the standard deviation of the sampling distribution of a statistic. It indicates how closely the population mean is likely to be estimated by the sample mean.
The standard error of the mean (SEM) is the standard deviation of the sample-mean’s estimate of a population mean
The SE is different from Standard Deviation (SD) which measures the amount of variability in the population
To Calculate SEM:
SEM = SD / square root of sample size
Confidence Intervals
Calculation of 95% CI for mean = (mean - 1.96xSEM) to (mean + 1.96xSEM)
ROC curve
Compares sensitivity and specificity
Y axis Sensitivity (True positive rate)
X axis 1-specificity (False positive rate)
Allows you to find the optimum cut off point
Better top left hand corner
closer to this - the more accurate the test // more area under curve = better
Can find out TP/TN/FP/FN from this if given prevelance
Validity
Accuracy of a test
How true to actual population
To increase - reduce systematic error
(TP+TN)/ (all - TP+TN+FP+FN)
Reliability
AKA Precision
How close the results are to each other
To increase - Reduce random error
Relative risk
A measure of the chance of the event occurring in the experimental group relative to it occurring in the control group.
e.g. your risk is increased by 10 fold
Relative risk (RR) is: a ratio of proportions.
RR = EER/ CER
EER = Experimental event rate =
Experimental group with outcome/ all experiemental group (a/a+b, smokers with cancer/ all smokers)
CER = Control event rate =
Control group with outcome/ all Control group (c/c+d, non-smokers with cancer/ all non-smokers)
COHORT
Relative risk reduction
RRR = (CER - EER)/CER
RRR = 1-RR
Use if the event is bad
Note: termed Relative benefit increase if the event is good e.g. smoking cessation.
Absolute risk reduction
The absolute difference between the risk of the event in the control and experimental groups.
ARR = CER - EER.
ARR can be used to calculate the number needed to treat (NNT).
Use this term if the event is bad e.g. death
Note: termed Absolute benefit increase if event is good e..g smoking cessation
Number needed to treat.
The number of patients who needed to be treated to prevent the occurrence of one adverse event (e.g. complication, death) or promote the
occurrence of one beneficial event (e.g. cessation of smoking).
NNT = 1/ARR.
Odds
Odds is: a ratio.
Odds of event in expt. group = a/b
Odds of event in control group = c/d
Odds ratio (OR) is: a ratio of ratios.
OR = ad/bc
e.g. a person is 10 x more likely to have (NOT your risk is increased by this number
CASE CONTROL
Prevalence
The proportion of existing cases
Incidence
The ratio of new cases to those at risk of developing the disease
no of people with disease/ (total no - no of people with the disease) - rate of new cases
Relationship between incidence and prevalence
Prevelance/ (1- prevelance) = incidence rate x disease duration
As disease duration becomes low, becomes more similar to prevelance number.
Chronic diseases, prevalence = higher than incidence
Levels of evidence
Ia - Evidence from Meta-analysis of Randomized Controlled Trials
Ib - Evidence from at least one Randomized Controlled Trial
IIa - Evidence from at least one well designed controlled trial which is not randomized
IIb - Evidence from at least one well designed experimental trial
III - Evidence from case, correlation, and comparative studies.
IV - Evidence from a panel of experts
Paired vs unpaired data
Paired data - same individuals, different time points. Found in longitudinal studies or case control.
Unpaired data - groups of data with different subjects, must not be influenced or related to another group.