Statistics Flashcards
What is linear regression?
A model that assumes a linear relationship between x (independent) & y (dependent) variables. Before attempting a linear model, should determine if there is a relationship between the variables, typically by use of a scatterplot.
+How do sens & spec differ cf +ve & -ve predictive values?
sens & spec refer to the characteristics of the TEST, while PPV & NPV assist with the clinical relevance of the test. They relate to prevalence; as prevalence reduces, PPV will reduce but NPV will increase.
+ What’s sens? spec?
cf PPV & NPV?
sens= the proportion of people who test +ve among those who have the disease
spec= the proportion of people who test -ve among those who do not have the disease
PPV is the probability that with a +ve test result the person will actually have the disease
NPV is the prob that with a -ve test result the person will not have the disease
:)why is statistical testing undertaken?
to deal with the problem of random variation in samples & to predict how likely it is that the results of the sample accurately reflect what we’d see in the entire population.
:) what’s the null hypothesis?
the null hypothesis is assumed until a level of statistical evidence is reached allowing us to reject it. Generally, the null hypothesis is that the exposure or intervention being studied is NOT associated with the outcome of interest.
:) what are possible explanations for the outcome of a study?
Truth: conclusion accurately reflects the answer to the question.
Bias: there were errors in study design distorting results & impacting conclusions.
Confounding: one or more variables associated with both exposure & outcome of interest impacting study results
Chance: random variations in the sample of the population being studies led to erroneous conclusions. Type 1 error= random chance leading to a mistaken conclusion that there was an effect (alpha error). Type 2 error = random chance leading to a mistaken conclusion that there was no effect (beta error)
:) what’s a p value?
a p value is a measure of the effect of chance within a study, the probability that if the null hypothesis were true, and if the results were not affected by bias or confounding, that we would see a result as extreme or more extreme than the one seen in the study.
:) what’s risk reduction?
refers to absolute or relative risk reduction.
absolute risk is the number of events in the treated or control group/ the number of people in that group
Absolute risk reduction= (AR of events in control group) - (AR of events in treatment group)
relative risk (aka risk ratio) is the AR of events in the treatment group / AR of events in control group
relative risk reduction= (ARC-ART)/ARC or 1-RR
NNT is 1/ARR
:) what are confidence intervals?
the range of values that could be considered reasonably likely after statistical tests are performed on gathered data; in correctly performed studies, we expect the confidence interval to surround the true value for the outcome (eg. risk reduction) 95% of the time.
more narrow confidence intervals= higher precision
confidence interval crossing 1= no difference between arms of a study
:) what are odds ratios? when are they useful?
represent the probability of event occurring in the treatment versus control group, ie. if A= disease in exposed, C= disease in unexposed, B=no disease in exposed, D=no disease in unexposed, odds ratio= (A/C)/(B/D)= AD/CB
overestimate the relative risk for outcomes which are common (ie. if the OR is >1, it is larger than the true relative risk, if it’s <1 it’s smaller). OR (or RR) of 1 means no association between outcome & exposure
should use OR for case control or retrospective studies, RR for prospective
:) what’s NNT?
the number of patients that need to be treated with an intervention to prevent a bad outcome. 1/ARR
AR_SC 2.1
Describe the basic concepts of evidence-based medicine, including levels of evidence, meta-analysis and systematic review
Describe the limitations of evidence-based medicine
EBM= the care of patients using the best available research to guide clinical decision-making.
Evidence may come from primary research; data collected from individuals or groups. For studies evaluating therapy or harm, RCTs are superior to obs studies which are superior to unsystematic clinical observations. such questions (benefits & harms) are best answered w RCTs. Questions regarding risk factors for disease & prognosis are best answered w prospective cohort studies
To limit bias, consider LEVELS OF EVIDENCE HIERARCHY:
I= systematic R/V of all relevant RCTs (meta-analyses higher)
II= at least 1 well-designed RCT
III-1: well-designed pseudo-randomised
III-2: comparative studies w concurrent control but no randomisation cohort studies, case-control
III-3: comparative studies w historical control or single arm studies
IV: case series
V: case reports
systematic reviews are best for answering single questions. scientifically structured, explicit about how relevant articles were collected, how scientific quality of each study was judged, weight of evidence from multiple studies. aim to avoid publication bias by aiming to include all strong research whether published or not (pub bias: +ve studies tend to be published).
Meta-analyses (often included in systematic reviews), is statistical method of quantitatively combining or pooling results from different studies. Can be used to provide overall pooled effect estimates.
SR & MA benefits= comprehensive assessment of the body of knowledge.
Overcome the limitation of generalisability of RCTS involving single populations/centres. incr sample size provides more precise estimates of effect size (ie. estimates that have smaller CIs). can allow exploration of heterogeneity. Unlike many narrative R/Vs, SRs start with a clinical or research questions & form conclusions based on evidence (vs starting w conclusion).
Given that they synthesise large body of evidence, transparent/explicit approach, consider strengths/weaknesses of individual studies/populations & interventions/outcomes, they can be used to determine best pt management decisons & inform guidelines.
Despite advances in research methods, many published study results are false or draw misleading conclusions.
some questions are not possible to obtain robust answers with adequate sample size & ethics.
adult learning
Adults have higher sense of self-direction, they use their life experience to facilitate learning & need to know how the info is relevant (therefore problem-based useful).
They want to choose how they learn & broadly apply different learning styles that work for them:
visual
auditory
read/write
kinaesthetic (tactile)
Challenge= some information is best suited to certain formats (eg. read/write)
adults have other distractions/responsibilities in their lives
opportunities/how to enhance= blend of clinical experience (visual/auditory), didactic lectures, clinical skills lab sessions (kinaesthetic)
adults seek continuous learning baed on personal interests/wants/needs (they are more intrinsically motivated but need to see the relevance of what doing/learning)
mentors useful
allow adults to organise themselves & work things out for themselves
project-based, problem solving, application into real world
