Thread by @EmilyCSparks_MD

Step 2 HY biostats explained 🧵 #MedTwitter #MedStudentTwitter #Step2

1) know how to draw a 2x2 table. Disease/Outcome is always at the top and the Test/Exposure is on the left-hand side

sensitivity (TP rate - those WITH the disease who actually test +) = (TP/TP+FN)
specificity (TN rate - those without the disease who actually test -) = (TN/TN+FP)

1a) A highly SENSITIVE test can RULE OUT (SNout) a disease if negative, and a highly specific (Spin In) test can RULE IN a disease if positive
- use high sensitivity test for SCREENING
- use high specificity test to CONFIRM

2) with a 2x2 table you can also calculate
PPV (# who test + for the disease that HAVE the disease) = TP/TP+FP; also, PPV ⬆️ with ⬆️ PREVALANCE
NPV (# who test - for the disease who DON'T have the disease) = TN/TN+FN

3) Likelihood ratios (LRs) - utility of a test - used to grade clinical significance of various results when >2 different test results are possible

LR- = 1-sensitivity/specificity; if LR- < 0.1 RULE OUT
LR+= sensitivit/1-specificity; if LR+ >10 CONFIRM DZ

4) p value ≤ a value = statistically significant @ that significance level & null hypothesis rejected

null hypothesis = statement of no relationship between the exposure and the outcome

type 1 error (a) - null rejected when true (false +)
type 2 error (β) - false negative

4a)
Power = 1 - β = simplistically, power is the chance of NOT making a type II error (β).
⬆️power by ⬆️sample size (this ⬇️confidence interval)

5) Relative Risk (RR) - used in cohort studies; ratio of the probability of an outcome occurring in the exposed group compared to it occurring in the unexposed

=Risk(tx)/Risk(control) OR AD/BC

RR>1 = exposure ⬆️ risk outcome
RR<1 = exposure ⬇️ risk outcome

5a) Absolute Risk (AR) = absolute difference between the risk of an outcome occurring in exposed individuals and unexposed individuals
= Incidence Risk(exposed) - Incidence Risk (unexposed)

6) Relative Risk Reduction: to determine how much the treatment reduces the risk of negative outcomes

=(Risk(unexposed)-Risk(exposed))/Risk(unexposed)

RRR=1-RR where RR=Risk(exposed)/Risk(unexposed)

7) vs Absolute Risk Reduction (ARR; risk difference)
difference btwn exposed group and nonexposed after INTERVENTION

eg, risk of death - Measures how effective an intervention was on a specific outcome

(absolute risk(unexposed ) - (absolute risk(exposed) = c/(c + d) – a/(a + b)

8) Number needed to treat (NNT) = The # of individuals that must be treated, in a particular time period, for ONE person to benefit from treatment (i.e., to not develop the disease)
=1/ARR

correlates with effectiveness

9) Number Needed to Harm (NNH) = # individuals who need to be exposed to a certain risk factor before ONE person develops an outcome
=1/ARI
ARI = Incidence Rate(adverse event, drug)-Incidence Rate(adverse event, placebo/control)

correlates with SAFETY

10) Temporality of study designs:
Cohort studies - retrospective or prospective
Case-control - study if an exposure (i.e., a risk factor) is associated with an outcome (i.e., disease) IN THE PAST
Cross-sectional - NOW: risk factor + or - ➡️ dz prevalance

11) other studies:
crossover: an experimental study design in which each participant switches from the intervention group to the control group or vice versa with a washout period in between
⬇️risk confounding bc each person serves as their own control

12) Factorial design (w/ random example)
= Multiple interventions are studied simultaneously by assigning participants to various combinations of interventions and placebo (or randomization to different interventions with additional study of two or more variables)

13) intention to treat analysis vs per-protocol analysis:
intention-to-treat analyzes as randomized (once randomized, always analyzed).
per-protocol = DROPS PATIENTS FROM ANALYSIS WHO DIDN'T FOLLOW PROTOCOL

14) observational biases
-recall bias
-observer bias
-reporting bias
-surveillance (detection) bias

15) Selection biases:
-ascertainment (sampling) bias - study population isn't the same as the target population bc of nonrandom selection methods used
-nonresponse bias
-Berkson bias - study disease in HOSPITAL PATIENTS
-Prevalence bias - exposure b4 dz assessment
-Attrition bias

16) EFFECT modification - phenomenon in which an internal factor modifies the effect of a RF on the outcome of interest
CONFOUNDING - A type of systematic error in which a 3rd variable that has not been factored into the study affects the independent and dependent variable

17) Cutoffs:
DECREASE: ⬆️ sensitivity ⬇️specificity ⬆️NPV & ⬇️PPV
INCREASE: vice versa

18)
LEAD time bias - NO ACTUAL DELAY IN MORTALITY - survival time is overestimated bc of EARLY diagnosis thru screening
LENGTH time bias - survival time is overestimated bc screening tests have a ⬆️ probability of detecting slowly progressive cases (longer asymptomatic phase)

19) internal & external validity (accuracy)
internal - extent to which study is free of error (mainly biases) & results are true for sample study; ⬆️ internal validity - control for age, sex, etc., refine measurement to ⬇️ systematic errors (bias again)

cont.

20) external validity - extent to which study results can be extrapolated from a sample population to the GENERAL population (generalizability)
-⬆️external validity - study results can be reproduced in diff sample groups, ⬆️ internal validity

21) precision (reliability) - is the test reproducible with the same results on the same sample under similar conditions?
-⬆️precision has ⬇️random error, precision improves with ⬇️standard deviation and ⬆️power of a test

22) more on study types:

Case-control: you get 2 groups of people with SIMILAR CHARACTERISTICS. Group 1 has the dz. Group 2 does not.
You ask about EXPOSURES they may have had (**recall bias**)

Know it’s associated with odds ratioss

22a) if they’re mentioning an intervention (eg, drug given) it’s not a case-control study or cohort study. They both deal with EXPOSURES.

RCTs deal with interventions.