Term
1. Sensitivity (SN) 2. Specificity (SP) 3. Type I error 4. Type II error 5. α/P-value 6. β 7. (Statistical) Power |
|
Definition
1. Probability that a person with the disease will test + SN=TP/(TP+FN) 2. Probability that a person without the disease will test - SP=TN/(TN+FP) 3,5. Rate of obtaining + test in absence of disease; "rejecting the null hypothesis when it is in fact true" α=p=FP rate=FP/(FP+TN)=1-SP 4,6. Rate of obtaining - test with disease; "accepting the null hypothesis when it is in fact false" β=FN rate=FN/(FN+TP)=1-SN=1-Power 7. Power=probability that a test will reject a FALSE null hypothesis/NOT make a Type II/β error
|
|
|
Term
Preventitive Medicine: 1. Primary prevention 2. Secondary prevention 3. Tertiary prevention CC example: breast cancer |
|
Definition
1. disease prevention 2. disease early detection to prevent progression or reduce severity
3. tx/rehab once disease has occured to minimize effects or recurrence CC example: 1. have kids early & often, BRCA screen, low fat diet 2. mammography, clinical breast exams 3. counseling/coping, tamoxifen, aromatase inhibitors, lymphedema prevention |
|
|
Term
7 criteria for an effective/worthwhile screening test: |
|
Definition
1. has to be bad (morbidity/mortality) 2. sufficient prevalence 3. acceptable SN & SP of test 4. safety of patient
5. minimal FPs (work-up=minimal harm)
6. intervention must be beneficial (treatable?)
7. cost-effective test
|
|
|
Term
Note on SN & SP significance vs. LRs |
|
Definition
SP & SN are characteristic of the test itself Likelihood given test results depends on BOTH SN & SP of the test AS WELL AS the prevalence of the disease!
|
|
|
Term
Predictive values: 1. PPV 2. NPV Likelihood ratios: 3. PLR 4. NLR |
|
Definition
1. PPV=probability of disease given + test result 2. NPV=probability of no disease given - test result *affected by prevalence*
ratio: means it is going to be a multiplicative factor 3. PLR=Times more likely a + test is obtained from a person with the disease compared to one without the disease PLR=SN/(1-SP) 4. NLR=Times more likely a - test is obtained in a person without the disease compared to one with the disease NLR=(1-SN)/SP |
|
|
Term
ROC curve: 1. What is it (stands for)? 2. What is plotted on the x- and y-axes? 3. What does an excellent test plot like? A shit test? 4. What does the slope of a tangent to the line represent? 5. What does the area under the curve (drawn vertically down to x-axis from a point chosen on the line) represent? |
|
Definition
1. ROC=receiver operating characteristic. It reveals the balance btw. SN and SP (inversely related). 2. y=SN (or TP rate; TPR), x=(1-SP) (or FP rate; FPR) 3. An excellent test will plot more like a 3rd or 4th-root equation (swoop up near upper L corner, remaining near to y-axis with an asymptote of SN=1.0. A 100% worthless test will be a straight line drawn from lower L-->upper R (corresponding to SN=0, 1-SP=0 in lower L, and SN=1, 1-SP=1 in upper R). 4. =PLR 5. =accuracy of test
|
|
|
Term
Types of bias: 1. Spectrum 2. Lead-time 3. Length-time 4. Work-up |
|
Definition
1. sample has higher prevalence or disease severity 2. test diagnoses disease earlier than old test, but only appears to increase survival as a result of this timeframe difference: patient dead at same time, just knew about it eariler 3. screening is more likely to catch diseases that are more drawn out in timeframe than ones that progress more rapidly 4. patients receiving +/- test results are preferrentially treated (receive "gold std. test") |
|
|
Term
Hierarchical model for Dx Technology Assessment: 1.
2. 3. 4. 5. 6. |
|
Definition
1. Reproducibility, image quality (ex. MRI), etc. 2. Acceptable SN & SP 3. Diagnostic change/impact: does it help in Dx? (Ratio: post-test odds : pre-test odds) 4. Therapeutic impact from test (treatable?) 5. Patient outcome impact; patient happier/improved? 6. Societal impact (healthcare costs, CEA in societal perspective?) |
|
|
Term
Data types: 1. Continuous 2. Dichotomous 3. Discrete 4. Ordinal 5. Nominal 6. Interval |
|
Definition
1. Data could be any value on God's green earth 2. 2 mutually exclusive options (lived vs. died) 3. An integer value (set # answers) 4. There is a hierarchy, but not necessarily the same degree of separation btw. 2 contiguous values 5. Data grouped (race, sex, etc.) but not ordered 6. Size difference btw. values is set (temp., etc.) |
|
|
Term
Differences btw. related terms: 1. Parameter vs. statistic 2. Patient-oriented vs. disease-oriented outcomes 3. Foreground vs. background questions
|
|
Definition
1. Parameters refer to the population, statistics refer to the sample group 2. Results of the study relating to patients (improved lives, mortality, etc.) vs. outcomes that change an aspect of the disease (LDL, etc.) but without improvement in mortality/morbidity 3. Qs that are patient-specific (for dx, clinical decisions on care), vs. Qs that are background (textbook/general info.). |
|
|
Term
Types of studies: Clinical: Observational: 1. Cohort 2. Case-control 3. Cross-sectional |
|
Definition
Clinical/experimental: focused, designed to test 2 different treatments (prospective) Observational: 1. Cohort: 2 or more groups chosen, one of which is "exposed" to some factor. Outcome=incidence of disease. Can be retrospective, can test for multiple outcomes. 2. Case-control: START with 1 outcome (disease), look for "exposure". Retrospective, outcome=odds ratio (for exposure(s)) since incidence cannot be calculated. 3. Cross-sectional: measures BOTH exposure & outcome at a point in time (ex: survey). No follow-up.
|
|
|
Term
Just definitions: 1. Survival/Kaplan-Meier analysis 2. Hazard ratio 3. Intention-to-treat analysis 4. Allocation concealment 5. Effect size |
|
Definition
1. Accounting for death/termination of subjects in a prospective study from one time interval to the next 2. Ratio: predictor/"avg. predictor">1=assoc. with decreased survival (<1=incr. survival, =1 no assoc.) 3. ALWAYS keep participants in the SAME GROUP from start to end of study, even if they jump groups; effectiveness measure more realistic. 4. Researchers do not know which groups they are putting the enlisted subjects into (opaque envelope) 5. Expected difference in outcomes btw. exp. & ctrl. groups |
|
|
Term
Some terminology distinctions: 1. Risk vs. relative risk (RR) vs. relative & absolute risk reduction (RRR vs. ARR)
2. Number-needed-to-treat (NNT) vs. NNH (harm) 3. Probablilty-->odds, odds vs. odds ratio (example) |
|
Definition
1. Risk=1 in 1000 (.001/yr), relative risk=risk + treatment/risk without: 1/10,000 / 1/1000=0.1, so RR=0.1 (10%) if treatment reduces risk to 1 in 10,000. ARR=risk exposed - risk not (or risk + tx - risk w/o tx)= RRR=(risk exposed - risk not)/risk not Note: reduction=these are always (+) values! 2. NNT=# patients that would have to receive treatment to prevent at least 1 bad outcome. NNH=# patients that would need to be exposed to produce an outcome=disease 2. odds=prob/(1-prob). Odds ratio= odds1/odds2 (kind of a multiplicative effect) Odds of rain today=80%, tomorrow=20% (.8/.2)/(.2/.8)=(.8 x .8)/(.2 x .2)=.64/.04=16x more likely to rain today 90% men drink, while 20% women do...odds of men drinking are 36x more likely (NOT 4.5x). |
|
|
Term
More distinguishing btw. bias types: 1. selection 2. recall 3. misclassification 4. measurement |
|
Definition
1. one or more groups not representative of the target population; exclusion criteria problem maybe
2. (in case-ctrl.) those affected more likely to recall having problems than those not 3. assignment to wrong groups (if non-random only) 4. results not measured in the same way btw. experimental and control groups |
|
|
