Exam 2 Flashcards
Hazard ratio 95% CI
Should not include 1
How do you get a Hazard ratio?
Cox Proportional Hazard Regression
Time to event (survival) outcome
Dichotomous independent variable
Clinical questions answered by survival analysis
Estimate time-to-event for a group of individuals
Compare time to event between two or more groups
Assess the relationship of co-variables in time to event
Why use survival analysis?
Using a t-test or linear regression ignores censoring (the time people spent in the study before drop out).
Same reason as to why we don’t use risk/odds ratios or logistic regressions
Censoring
Subjects are said to be censored if they are lost to follow up or drop out of the study or if the study ends before they die or have the outcome of interest. They are counted as alive and disease-free for the time they were enrolled in the study.
This is why survival analysis is so popular.
What is survival analysis?
Models time to failure or time to event.
-Unlike linear regression, it has a dichotomous (binary) outcome.
Able to account for censoring
Can compare survival between 2+ groups
Assess relationship between covariates and survival time.
How do you report survival analysis?
Kaplan-Meier curve Median survival time Restricted mean survival time (RMST) Measures in follow-up time Hazard Ratios
Kaplan-Meier curve
Horizontal axis: ALWAYS time
Vertical axis: A measure of survival
Each drop in curve represents that at that point in time an event happened.
Many graphics are truncated (do not show 0)
Vertical lines represent censored observations. This gives you an idea of when you started losing patients.
Dependence Scale
Time to event analysis
An easier to read kaplan-meier curve
Assesses 6 levels of functional dependence. Time to event is the time loss of 1 dependence level (increase in dependence)
Key points for Kaplan-meier curves
Make sure to read the axis labels carefully
See the marks and labels on the horizontal axis
Labels should represent MEANINGFUL time points
-Example- yearly for a 5 year study and every 5 years for a 20 year study
Survival curves are less precise at the end because of fewer subjects.
Hypothesis testing in survival analysis
H0= S1(t)= S2(t)
Time to event outcome, dichotomous predictors
Analysis: log rank test
Reject the null if p-value is less than 0.05
When critiquing survival analysis articles:
Look at censoring- there should be a description of how censoring arose.
Beware of comparison of rates at the end of the study.
Median survival time
The estimated time point at which 50% of the study population has experienced the event (or still alive) or time when probability of surviving is 0.5.
Median time to the primary endpoint.
What happens if >50% of the study population has not experienced the event at end of study?
Report 25% survival time instead
Restricted Mean Survival Time (RMST)
The calculation of the MST requires the estimation of the entire survival function.
This is almost impossible due to drop out and limited f/u.
This is why RSMT is so popular. It is the estimate of the mean survival times in a restricted (truncated) time interval.
“Among those in the study who had an event within the first 24 years, the average time (mean) to event is 20.1 years.”
Measures of follow-up time
Incomplete and differential follow-up between the primary comparison groups can lead to informative censoring and bias and bias may be overlooked if no summary measure of f/u
Hazard ratio is used for what?
To adjust for predictors/ confounders and censoring.
Interpret as any other rate ratio.
The point estimate indicates that at any point tin time the hazard of (outcome) for the intervention is (XX) times the hazard for the comparison group.
Intersection in kaplan-meier curves
When 2 curves never intersect they are likely to be statistically significant
Systematic Review
A systematic review is a clearly formulated question that uses systematic and explicit methods to identify, select and critically appraise relevant research, and to collect and summarize the data from the included studies.
Key characteristics of a systematic review
Objectives- clear, pre-defined
Methodology- Explicit and reproducible
Comprehensive-Attempts to identify all studies that would meet the eligibility criteria
Validity
What is a meta analysis?
- ) Answers a focused clinical question
- ) Searches the medical literature
- ) Assesses studies for inclusion and quality
- ) Mathematically combines the results of studies to determine a statistical summary statistic.
Medical analysis goals
Resolve clinical controversy
Provide a summary statistic
Increase precision (narrow CI) around a summary statistic by increasing ss
When are meta-analyses useful?
Quantitative and objective assessment
Useful when the results of previous trials are inconclusive or contradictory
Useful when ss is small
However- the quality of a meta analysis depends on the quality of the studies in it.
Unlike systematic reviews, practice guidelines use what?
Consensus opinion when evidence is insufficient.
Evidence based medicine steps
Ask and appropriate answerable question
Find the evidence- PubMed/filter
Apprise the evidence- validity, statistical sig, etc.
Apply evidence to practice.
Apprising the evidence of a systematic review
1.) What question did the systematic review/MA address?
It should be clear and focused
It should describe the population, intervention/exposure, and outcomes of interest.
2.) Look to see if a comprehensive search of all relevant studies (published and unpublished) were identified. Search strategy should include both controlled vocab terms and text words
Were the criteria used to select articles for inclusion predetermined, clearly stated, and appropriate? Inclusion and exclusion criteria should be clearly defined.
3.) Assess publication bias
4.) Determine validity
5.) Were the results similar from study to study?
-Results should be homogenous
Funnel plot
A method to visually inspect the studies in a MA and determine if publication bias exists.
Y axis- sample size
X axis- summary statistic (odds ratio)
The results of large studies show lower variability and are clustered together at the top of the funnel. The results of small studies show greater variability and are spread out at the bottom of the funnel.
This inverted funnel shape means that publication bias was unlikely.
Statistical test assessing publication bias
Beggs test
p<0.05 is significant for publication bias
How do you determine validity of a MA?
RCTs- Jadad score, Cochrane risk of bias
Diagnostic studies- modified QUADAS checklist
Reporting results in a systematic review
Report study characteristics- study size, f/u, PICOs, and include citations
Risk of bias w/in studies
Results in individual studies
Reporting results in a meta analysis
Everything from systematic review PLUS
Forest plot
Risk of bias
Additional analysis
Forest plot
Graphically assess the heterogeneity of study results
X axis- summary statistic
Y axis- Odds ratio of one. An odds ratio of one indicates no difference. If the forest plot has a mean the line of no effect is 0.
Results to the right of 1 favor the intervention and those to the left favor control/
Forest plot- heterogeneity
If no heterogeneity - Similar in magnitude -Similar in direction of effect - Overlapping CIs If these are present the studies can be combined.
What do we do if 1 study represents heterogeneity?
Propose possible subgroups pre-analysis
Re-examine study
Consider sensitivity analysis- summary statistic with/w/o Study
Statistical heterogeneity
Amount of variation in treatment effect present in MA beyond chance
Examined and quantified using statistical tests.
Best choice for MA
Cochrane Q test (chi-square test)- significant p value =heterogeneity. Only tells you yes/no.
I^2 value
I^2 value
Percentage of total variation across studies that is attributable to heterogeneity rather than chance. Ranges between 0 and 100% Typically, if I^2 <25%= homogenous 50% moderate >75%= high heterogeneity
Goals of CPGs
Improve quality of patient care Reduce unnecessary variation in practice Lessen disparities in practice Empower patients Influence public policy.
Who develops CPGs?
Governmental organizations
Professional organizations
CPG developmental process
No clearly defined “gold standard” method
Generally:
Identify topic and group of developers
Systematic search for evidence related to topic
Evaluation and determination of strength of evidence
Develop recommendation
update
National Academy of medicine provides standards.
National Academy of Medicine CPG standards
- ) Establish transparency
- ) Manage conflict of interest
- ) Guideline development group composition
- ) CPG-systematic review intersection
- ) Establish evidence foundations for an rating strength of recommendations
- ) Articulation of recommendations
- ) External review
- ) Updating
How to manage COI in CPG
Simple disclosure Exclusion from leadership roles Participation in certain restricted recommendations Formal or informal consultation Full exclusion
CPG- composition of guideline development group (GDG)
GDG must b e multidisciplinary and balanced
Should include patient and a patient advocate or a patient/consumer org rep
CPG-systematic review intersection
GDG should use systematic reviews that meet the gold standards of IOM.
GDG and systematic review teams should interact.
GRADE
High quality- multiple RCTs or high quality meta analyses
Moderate Quality- 1-2 RCTs or mod quality meat analyses
Low quality
Very low quality
Strength of recommendation
Considers
Balance between desirable vs undesirable outcomes
Quality of evidence
Uncertainty or variability in values and preferences
Strong recommendation- implies most pts should receive intervention.
Recommendation structure
Pt population
Baseline risk of pop
Quality of evidence
Strength of recc
CPG- external review
Reviewers should comprise a full spectrum of stake holders
Authorship kept confidential
GDG considers ALL comments for modifying or not modifying
Draft available for public comment
CPG- updating
CPG publication date Date of systematic evidence review Proposed date for future review Literature should be monitored CPGs should be updated when NEW EVIDENCE suggests need.
Assessing quality of CPGs
AGREE- to be used as a group and scores compiled and compared. heavily influenced by extent to which developers describe their process.
iCAHE- individual use
What do CPGs look like?
Preamble Introduction Summary of revisions Clinical questions or topics Limitations/future research Disclosures
Quality Improvement Projects
Strong recc imply that benefits outweigh harm, feasibility, and acceptability.
These are candidates for QIPs.
Concerns with CPGs
Application of multiple guidelines to one patient
Heterogenous patient populations
Implications for lack of adherence to guidelines.
Pharmacoepidemiology
Study of the use of and the effects of drugs in large numbers of people
PV
The science and activities relating to the detection, assessment, understanding and prevention of AE or any other possible drug related problems
Problems with clinical trials
Expensive Small Drugs compared against placebo Excludes elderly, children, pregnant pts, comorbidities Too short to detect long term AE Too small to detect rare AE May be unethical
Contributions of pharmacoepi
Information which supplements the information available from premarketing studies- better quantification of the incidence of known AE and beneficial effects
New types of info not available from premarketing studies
AE
Any unexpected medical occurrence in a patient administered a medicinal product which does not necessarily have to have a causal relationship with this treatment
ADR
All noxiuos and unintended responses to a medicineal product dose related
PV methods
Review case reports and case series from spontaneous reporting systems
Variety of types of observational studies
Spontaneous reporting systems
Core aspect of PV
Pharmaceutical manufacturers have to report- FDA MedWatch
There is significant underreporting.
AEs/ADRs for which the relationship to a drug can be established
A low or near absent frequency in the underlying pop
Are not part of the underlying illness being treated
Are generally the result of exposure to a drug toxin and have no other likely explanation
Important regulation requirement- PV
Pharma must apply causality assessment for events associated with their drugs for regulatory compliance.
Sponsors must report serious unexpected events in clinical trials where there is a “reasonable possibility” that the events were caused by the drug
No admission of causation
Use-
unstructured clinical judgement/ global introspection (very subjective)
Algorithm/criteria method with verbal judgments (yes/no, simple, improves consistency)-
Naranjo scale
Probablistic methods
PV- reporting ratios
Incidence can not be calculated with spontaneous reports
Incidence= #events/ #ppl exposed
Data sources for PV
Spontaneous reporting systems -FDA MedWatch -New gen- Social media Automated databases -Claims and other administrative databases -Medical records
Claims and other administrative databases
Pros (pharmacy claims)- best data on drug exposure in PE
Cons (medical claims)- issues with coding, reimbursement does not usually depend on diagnosis.
Examples - Medicare Part D, Medicaid Claims, VA claims, private insurance
Medical records pros/cons
Pros: Validity of the diagnosis values
Labs values
Cons- Incompleteness of patient data (fragmented health systems)
Use of automated databases
- ) Looking for uncommon outcomes because of a large sample size
- ) A denominator is needed to calculate incidence rates
- ) One is studying short term drug effects
- ) One is studying objective, lab driven diagnosis
- Bias could influence association
- ) Time and budget are limited.
Case reports
Events observed in single patients
Raises hypotheses about drug effects, to be tested with more rigorous study designs.
Case reports limitations
N=1
Cannot know if the patient reported is either typical of those with the exposure or typical of those with the disease.
Cannot usually determine whether the AE was due to drug exposure
Very rare that case reports are used to make a statement about causation, BUT when the outcome is so rare and so characteristc of exposure it is considered likely due to it.
Case reports causation
When the disease course is very predictable and the treatment causes a clearly apparent change in the disease course.
Case series
Collections of patients, all whom have a single exposure, whose clinical outcomes are then evaluated and described
Case series can be collections of patients with a single outcome, looking at their antecedent exposure
Limitation- no control
Ecologic studies- analyses of secular trends
Examine trends in an exposure that is a presumed cause and trends in a disease that is a presumed effect
Test whether the trends coincide.
These trends can be examined over tiem or across geographical boundaries.
Useful for rapidly providing evidence for or against a hypothesis.
These studies lack individual data and only use aggregated group data. You are unable to control for confounders.
Observational Studies
No intervention- just looking at people already taking the drug. We are observers.
Observing with comparison group- analytical
descriptive0 no comparison group
Analytical observational studies
Cohort
Case-control
Cross sectional
Case report
Simple description of observations in regards to a single pt
Case series
Simple description of observations in regards to group patients
Cross-sectional studies
Survey
- Provides data on a group of subjects at a particular time.
- Can be descriptive or analytical
- Descriptive- observing a population and putting which pts have or dont have outcome
- Analytical- they look at a population with a characteristic and those that do not for looking at an outcome
Prevalence
# of people with a disease at a point in time compared to the # of individual sat risk for disease' Ratio
Incidence
The number of new cases of disease in a pop within a period of time
Rate
Case control studies
Observational, analytical
Group of patients with specific outcome- looking into their past to identify risk factors.
Case and control
These studies are RESTROSPECTIVE ( you start with the outcome)
Answers the question “what happened”
Confounders
Age and sex are risk factors for a number of diseases (confounders)
Researchers may match cases to controls for certain characteristics
Matching helps eliminate the impact of confounders
Differentiating between case control adn case series
Both written after the fact
What was the author trying to achieve?
They wanted to describe a phenomenon- case series
They wanted to explain a phenomenon by looking at previous events- case control
Cohort study
Observational, analytical
A group of people with characteristic in common who remain as part of a group for an extended period of time.
Researchers select groups
Patients are placed into cohort based on if they have the risk factor or not.
Patients are followed to determine what the effect of the risk factors is
Direction of inquiry is forward- BUT can be prospective OR retrospective
Answers the question “What will happen”
Explain retrospective cohorts
Example:
Observation: Gentamicin use can cause AKI, statins were found to reduce this issue in rodents.
Looked at electronic medical records of all patients in hospital who got gentamicin after x amount of years. Split pts into 2 groups, did they come in on a statin or not? Compare kidney function in both groups after gentamicin use.
This study compared old data (retrospective) but looked forward at the outcome (forward moving).
Cross-sectional
What is happening right now?
Neither prospective or retrospective
Case-control studies are useful for
Rare conditions or conditions that dont manifest for many years
Why are experimental study designs stronger?
They start with one pop and then isolate the variable (intervention)
Controlled trials are far superior at proving causality.
Reporting results in observational studies
Cohort- event rate
Case-control- how likely is it that you have a risk factor previously
How do we report event rate?
Absolute risk
NNT
Relative risk
The odds ratio
The odds that a patient with the outcome had the risk factor compared to the odds that a person without the outcome had the risk factor
>1= x time more likely
1= same between groups
Internal Validity
How valid is the trial? How likely is it that our identified cause was what led to the effect?
Main threats- bias, random error (findings occurred by chance)
External validity
How generalizable is the study
Do they patients represent a typical pop? are the results applicable to practice?
Assessing an article for validity
Do not look at title and abstract
The majority of assessment should take place in the methods and results
Hypothesis: :Smoking is a risk factor for lung cancer
Case-control- Look at group with and without lung cancer? Which had h/o smoking?
Cohort- one group smokers, next group non smokers. What happened?
RCT- Find group of non smokers and giving 1 group cigarettes
Selection bias
Did the patients represent typical pop with disease?
Can you use it for your patients?
What outcomes were they studying?
Were they subjective or objective?
Who measured them?
Many studies with subjective outcomes will assess outcome by 3rd party
Detection bias
Certain pt characteristics make the assessing an outcome challenging.
(obese men and prostate cancer)
Attrition bias
Loss of pts to f/u at different rates for different reasons between groups.
This may mischaracterize the overall group at the end of the study