Flashcards in Day 3.1 Behavioral Science Deck (135):
Gp of people w disease vs Gp of ppl w/o disease
Retrospective (looks back in time)
Observational- looking at possible risk factors for dz (not nec a causal relationship)
Asks "What happened?"
What is the measurement for a case-control study?
Odds ration (OR = ad / bc)
E.g. pts w COPD had higher odds of a hx of smoking than pts w/o COPD
Compares Gp w risk factor to Gp w/o risk factor to determine if risk factor increases likelihood of dz
Prospective (forward in time)
Asks "What will happen?"
Clinical trials are cohorts.
Measurement for cohort study?
Relative Risk (RR = [a/(a+b)] / [c/(c+d)]
E.g. smokers have a higher risk of developing COPD than non-smokers
Assesses freq of dx (and related risk factors) at a particular pt in time.
Asks "What is happening now?"
Measurement for cross-sectional study?
Can show risk factor association, but doesn't show causality
Twin concordance study
Compares freq with which both monozygotic twin or both dizygotic twins develop a dz
Compares siblings raised by biologic vs by adoptive parents.
Measures heritability vs influence of env factors
What is meta-analysis?
Pools and integrates data from many studies to get an overall conclusion. Has greater statistical power.
"Highest echelon of clinical evidence"
Limitations: quality of individual studies; bias in study selection
Disease is on top, test is on side
pos/neg are horizontal
TP / [TP + FN]
Or, 1 - FN
What is sensitivity?
Proportion of all people with dz who test positive. (TP over all people w/ disease)
Also, the ability of a test to detect a dz when it is present.
SNOUT - SeNsitivity rules OUT dz
TN / [TN + FP]
Or, 1 - FP
What is specificity?
Proportion of all people w/o dz who test negative (TN over all pts w/o dz)
Also, the ability of a test to indicate non-dz when no dz is present.
SPIN = SPecificity rules IN
Are HIV tests sensitive or specific?
First test is ELISA - it's very Sensitive but has a high FP rate (SN rules OUT, so it's better to get everyone who might have it rather than miss someone who doesn't).
If positive ELISA could be TP or FP, so confirm with Western Blot, which is specific, and high FN rate.
SP rules IN. So use Western as confirmatory test.
Which should be more sensitive, screening tests, or confirmatory tests? Which should be more specific?
Screening = Sensitive (SN rules OUT- so need to get all the possible positives- even if they are FP)
Confirmatory/diagnosing tests should be specific
SP rules IN.
TP / [TP + FP]
What is PPV?
The proportion of all positive results that are accurate.
Probability that pt has dz if pt has a pos test result
TN / [TN + FN]
What is NPV?
Proportion of neg test results that are correct.
Probability that a pt does not have dz, given negative result
How does high prevalence affect PPV and NPV?
If there are a lot of pts w dz, the number of TP will go up, and the number of FN will also go up (bc more pts have it, and some of those will wrongly test negative).
Increased TP means increased PPV.
Increased FN means NPV will decrease.
How does low prevalence affect PPV and NPV?
If prevalence is low, then the FP will go up, and also to TN will go up.
Increased FP means decreased PPV
Increased TN means increased NPV
Will a test be more accurate in a population with high prevalence or low prevalence?
High prevalence means a high PPV, and low NPV, so the test will peform better w high prevalence. This is why you should only screen pts who need screening.
What is (point) prevalence?
total cases / total population at risk
this is at one point in time.
What is incidence?
NEW cases during a given period / pop at risk during that period.
Incidence is NEW incidents only.
Pts who already have it don't count.
How can you approximate prevalence using incidence?
Prevalance is approx incidence (new cases) x disease duration
For chronic dz, which is larger, prevalence or incidence?
Prevalence is bigger than incidence
Bc duration is longer. There are more overall cases than there are new cases, bc it lasts so long
For acute dz, which is larger, prevalence or incidence?
They are equal. Prevalence (everyone that has it) and incidence (new cases) are the same, since the dz doesn't last long enough for prevalence to be bigger. (duration is short)
e.g. common cold
Odds ratio eqn
OR = AD / BC
What is odds ratio?
Odds of having dz in exposed divided by odds of having dz in unexposed.
Use for case-control studies (retrospective)
Relative risk eqn
RR = [a / (a+b)] / [c / (c+d)]
What is the relative risk?
Probability or getting the dz in the exposed gp vs getting the dz in the unexposed gp.
For cohort studies (prospective)
Attributable risk eqn
AR = [a / (a+b)] - [c / (c+d)]
What is attributable risk
The difference (so subtract!) in risk bt exposed [a / (a+b)] and unexposed [c / (c+d)] groups
Or, the proportion of dz that is attributable to the exposure.
Ex smoking causes 1/3 of the cases of pneumonia
Absolute risk reduction eqn
ARR = [a / (a+b)] - [c / (c+d)]
Note: same as AR, but now chart has risk factor Reducer rather than risk factor.
What is absolute risk reduction?
Difference in risk reduction between treatment (exposed) vs placebo (unexposed).
Same as attributable risk, but now looking at a risk factor reducer (treatment) rather than a risk factor.
What is the NNT?
Remember, ARR is looking at risk factor Reducers (aka TREATments) So the NNT is related to this.
Number of pts who have to get treatment before the drug successfully treats one pt. Want it to be low.
What is the NNH?
AR looks at risk factors (HARMful things), so the NNH is related to the AR.
The number of pts who could have the risk factor before one pt was harmed. Want it to be high.
Fornicators: 30% have HPV
Non-fornicators: 5% have HPV
What is the attributable risk of fucking to getting HPV?
AR = Exposed - unexposed
30-5 = 25%
If 10,000 pts took a drug and 100 were saved by it, what is the NNT?
NNT = 100.
Need to treat 100 patients in order to save 1 patient
(10,000 drug / 100 saved = 100 drug / 1 saved)
What is precision?
Consistency and reproducibility of a test (reliability)
Absence of random variation in a test.
What is accuracy?
Trueness of test measurements (validity)
Do random error and systemic error reduce precision or accuracy?
Random error = reduced precision
Systemic error = reduced accuracy.
RR < 1
The dz is LESS likely to occur in the group
Or, the factor is protective
RR > 1
The dz is MORE likely to occur in the exposed group
Or, the factor is harmful.
RR = 1
The factor makes no difference
What is validity?
What is reliability?
How to reduce bias?
1. blind studies (dbl blind is better)
2. placebo responses
3. crossover studies (switch treatment/placebo halfway through; ea subj acts as own control)
What is bias?
One outcome is systematically favored over another. Systematic errors.
(Systematic error = reduced accuracy/validity)
Non-random assignment to study group
Berkson's bias: using hospitalized pts
Knowing presence of disorder alters the recall by pts
Subj not representative of general pop
So results can't be generalized to general pop
Info gathered at inappropriate time. This is a type of recall bias.
E.g. Giving a survey to study a fatal dz- only live pts can answer
Subj in diff gps not treated the same
eg Treatmt gp gets more attn, stimulating greater compliance
2 closely assocd factors- one factor has an effect on the other and distorts/confuses the effect of the 2nd factor
Early detection is confused with increased survival- happens with improved screening for dz
The natural hx of dz is not changed, but bc of early detection, survival time seems to be increased
Researcher's belief in efficacy of rx chgs the outcome of the rx
The group being studied changes its behavior bc they know they are being studied.
What is positive skew?
mean > median > mode.
Asymmetry w TAIL on right.
What is negative skew?
mean < median < mode
Asymm with TAIL on left
What measurement is least affected by outliers in the same?
In a normal gaussian distribution, where are the mean, median, and mode?
in the middle. they all equal each other.
note: in bimodal (2 humps), there are 2 modes, so two peaks.
What is the null hypothesis (H0)
Hypothesis of no difference (no assoc bt dz and risk factor)
What is the alternative hypothesis (H1)
There IS a difference (there's an assoc bt dz and risk factor)
Type I error (alpha)
Saying there IS a difference when there isn't one.
Mistakenly accepting experimental hypothesis and rejecting null.
What is the probability of making a Type I error?
probability = p
p is judged against alpha, a pre-determined level of significance (usu <.05)
What is a Type II error (beta)
Stating that there is NOT a difference where actually there is one.
Not rejecting the null when the null is false and should have been rejected
What is the probability of making a Type II error?
Setting a guilty man free- what type of error?
Type II (Beta)
Convicting an innocent man- what type of error?
Type I (alpha)
What is power?
The likelihood of rejecting a null hypothesis when it is false and should be rejected, or the likelihood of finding a difference when a difference does exist.
What happens to power when you increase the sample size?
Power in numbers!
Eqn for Power
What does power depend on?
1. total # of end points experienced by population (don't want study drop-outs)
2. Difference in compliance bt treatmt groups
3. Size of expected effect (easier to be wrong abt a 1 year life expectancy increase than abt a 15 year increase)
What is the standard error of the mean?
st dev / (sq rt of sample size)
sigma / sq rt of n
SEM decreases as n increases
Normal gaussian distributions at st dev of 1, 2, 3
1 st dev = 68% (34+34)
2 st dev = 95% (42.5+42.5)
3 st dev = 99.7% (~50+50)
also 1.645 st dev = 90%
What is confidence interval?
Likelihood that if you repeated a study, the repeat would fall in the same range as the original
Eqn for CI
CI = mean +/- Z(SEM)
Z = predetermined level of confidence that you want to attain. The Z for 95% is 2 (actually 1.96)
SEM = st dev / sq rt sample size
Don't forget to do + and - so get it on either side of the mean!
68% what is the Z?
99.7% what is the z?
90% what is the z?
95% what is the z?
2 (but actually 1.96)
If the 95% CI for a mean difference between 2 variable includes 0, what does this mean?
If it includes 0, it means that 0 is a possible answer for if the means are different. If the means can have zero difference, it means that there is no difference between the two treatments.
If the 95% CI for odds ration or relative risk includes 1, what does this mean?
If RR =1, no difference between the two groups.
So H0 is not rejected.
If the CI between 2 groups overlaps, what does this mean?
The groups are not significantly different.
99% what is the z?
checks differences bt the means of 2 groups
checks differences bt the means of 3 or more groups
x2 (Chi squared) test
checks differences bt 2 or more PERCENTAGES or proportions of categorical outcomes.
Unlike t-test and ANOVA, is NOT looking at means.
Disease prevention classifications
PDR: prevent, detect, reduce disability
primary prevention: prevent dz occurrence (vaccine)
secondary prevention: early detection (pap smear)
tertiary prevention: reduce disability from dz (chemo)- not really prevention of dz at this point, just prevention of effects.
r, measures correlation bt 2 variables
closer to 1 means closer correlation
r is from -1 to 1
-1 means inversely correlated
0 means not correlated either way
1 means positively correlated
Coefficient of determination = r^2 (r squared)
Correlation does not imply causation.
Prevention measures for diabetics
Serum glucose levels
urine microalbumin- proteinurea, indicative of diabetic renal dz
serum lipids (coronary artery dz)
foot exams (neuropathy, poor circ = injury)
dilated eye exam (retinopathy, cataracts, glaucoma)
influenza and pneumococcal vaccines
Prevention measures of high-risk sexual behavior
HIV and syphilis screening
HPV screen and vaccine
counsel on STDs and condom use
Prevention measures for smoking
Address quitting at each encounter
Avoid Vit A supplements (excess Vit A incrs lung cancer risk)
avoid OCP in women over 35 (incr DVT, thrombembolism)
abd US in males 65-75 to r/o AAA (abd aortic aneurysm)
influenza and pneumococcal vaccines
Prevention measures for drug use
HIV and TB testing
Prevention measures for alcoholism
Influenza, pnuemococcal vaccines
Prevention measures for overweight
Blood sugar tests for diabetes
Prevention measures for homeless, recent immigrant, inmates
HIV, Gonorrhea, Syphillis
A, B, C
Leading causes of death in infants (<1yo)
Short gestation (low birthweight)
Maternal complications of prego
Respi distress syndrome
Leading causes of death in 1-14yo
Leading causes of death 15-24yo
Leading causes of death 25-64
Leading causes of death 65+
Amt the insured pt has to pay to the insurance company, monthly.
Amt insured pt pays at time of service
eg $30 for clinic visit
$15 for meds
Yearly amt- amt that insured pt has to pay out of pocket before the insurance company will pay. Once deductable is met, insurance company pays everything else
Pts get procedures at end of year bc of this
What is the financial duty of the pt w insurance?
What is the financial duty of the insurance company?
Health care expenses beyond what the pt pays- beyond co-pay and beyond deductable
What is the risk the pt takes in buying insurance?
Paying more to insurance than what is received in medical care
What is the risk the insurance company has in providing insurance to the pt?
Paying more for medical care than they get from the patient.
What is the reward to the pt and to the insurance company?
Reward to pt: financial benefit if cost of medical care exceeds cost of insurance
Reward to insurance company: financial benefit if pt pays more than cost of medical care
Condition that pt already has that insurance will not cover (bc it's too financially risky for insurance to cover it)
Max amt that an insurance company will pay in life. e.g. 3 mil.
Group of healthcare providers that has agreed to a reduced payment in order to have access to a larger number of pts
Health Maintenance Organization
PCP is gateway to specialist care
Provider must be in-network for insurance to cover the cost.
Preferred Provider Org
No gatekeeper to specialist (can see w/o going to PCP)
Provider does not have to be in-network, but out-of-network is more expensive to pt
The list of meds that an insurance company will pay for
Utilization Management (UM)
Dept that evaluates appropriateness, necessity, and efficiency of healthcare services (w/in a hospital or clinic)
See if pt can leave earlier, if billing is maximized, etc
Resource-based Relative Value Scale
Scale that determines what a doc should be paid for a specific procedure (CPT code) or specific service in a certain region of the country- based on amt of work req'd to do procedure, regional practice expense, and regional malpractice expense.
RVU = relative value unit. Different procedures in diff locations have diff RVUs
MCO / Health Plan
Managed Care Org / Health Plan / Health Care Org
Org that tries to maximize quality/minimize cost.
Encourages pts/docs to choose less costly care, controls in-pt admissions and length of stay, emphasizes prevention
Options for pts w/o access to care
VA (federally funded)
City/County/State funded health networks (e.g. tax that goes toward indigent care in that area)
Federally funded teaching hospitals
Universal health care
4 ways physicians are compensated
Pay for performance
Payment in exchange for a specific service
eg surgical procedure (compensation based on RVU)
clinic visit (higher complexity = reimbursed more)
Fixed payment for a period of time or for a number of pts (a cap on time, pts), regardless of number of procedures
eg ER shift, minor emergency shift
Concierge practice (fixed fee one year, do anything pt needs in that year)
HMO- bank of pts, get paid flat rate to treat all of them, no matter how often/not they see you
Specified amt regardless of work performed
Pay for performance
Payment is increased if a physician meets pre-established targets
eg MCO/Health care org pays more for meeting vaccine targets, HbA1C <7.5
Federal programs, started by amendments to social security act
MedicarE - Elderly (65+, disability, or end stg renal dz needing dialysis)
MedicaiD - Destitute. federal/state assistance for very low income pts
Medicare Part A
hospital (anything in-pt)
Medicare Part B
Medicare Part C
Combo of A + B
Paid for by govt but run by private insurance companies
Medicare Part D
Stand-alone prescription drug coverage
Children's Health Insurance Program
Matching state and federal funding for child health coverage