1/72
Looks like no tags are added yet.
Nominal data
Categories with no order (sex, blood type); two categories = dichotomous
Ordinal data
Ranked categories with unequal spacing (pain scale, NYHA class, cancer stage)
Interval data
Equal intervals but NO true zero (Celsius temperature)
Ratio data
Equal intervals AND a true zero meaning "none" (heart rate, weight, dose, serum creatinine); most clinical measures
Is a 0–10 pain score continuous data?
No, it's ordinal (numbers don't automatically make data continuous)
Best measure of center for nominal / ordinal / normal continuous / skewed continuous data
Mode / median / mean / median
Mean vs median vs mode
Mean = average (pulled by outliers); median = middle value (resists outliers); mode = most frequent value
A paper reports median (IQR) instead of mean ± SD. What does that tell you?
The data were skewed (or ordinal)
Range
Maximum − minimum; uses only 2 values, so it's easily distorted
Interquartile range (IQR)
Q3 − Q1 = the middle 50% of data; outliers can't distort it; reported with the median
Standard deviation
The typical distance of values from the mean; reported with the mean; a small SD means the mean describes patients well
In a normal distribution, how do the mean, median, and mode compare?
All three are equal and sit at the center of a symmetric bell curve
% of data within 1, 2, 3, and 4 SD of the mean
About 68%, 95%, 99.7%, and 99.99%
Positive (right) skew
Tail extends right; a few very high values; mean > median (e.g., hospital cost, length of stay)
Negative (left) skew
Tail extends left; a few very low values; mean < median (e.g., scores on an easy exam)
What does an outlier do to the mean, SD, and median?
Pulls the mean toward it and inflates the SD; the median barely changes
Histogram
Shows the frequency of continuous data (x = value bins, y = count); used to check normal vs skewed; two peaks may mean two populations
Boxplot
Box = IQR, line inside = median, whiskers extend to 1.5 × IQR, dots beyond = outliers; the visual form of median (IQR)
Scatterplot
Plots two continuous variables against each other; shows whether a relationship exists and is linear; correlation is not causation
Incidence
NEW cases ÷ population at risk over a time period; measures the risk of developing a condition
Prevalence
ALL existing cases ÷ total population at a point in time; measures burden of disease
Why can prevalence rise even when incidence doesn't?
People survive longer with the disease (for example, because of better treatment)
2x2 table layout
a = treated + outcome; b = treated, no outcome; c = control + outcome; d = control, no outcome
Risk
Events ÷ total in the group
Relative risk (RR)
Risk in treated ÷ risk in control; RR < 1 = reduced risk, RR = 1 = no difference, RR > 1 = increased risk
Relative risk reduction (RRR)
1 − RR
Absolute risk reduction (ARR)
Risk in control − risk in treated (in percentage points)
Number needed to treat (NNT)
1 ÷ ARR; always round UP; a smaller NNT means a more effective treatment
Number needed to harm (NNH)
1 ÷ ARI (risk of harm in treated − risk in control); always round DOWN; a larger NNH means a safer treatment
When do you use NNH instead of NNT?
When the treatment INCREASES the risk of an outcome (a harm), giving a negative ARR
Odds vs risk
Odds = events ÷ non-events; risk = events ÷ total
Odds ratio (OR)
ad ÷ bc; used in case-control studies, where risk can't be calculated
How do you interpret an OR of 1.16?
The odds of the event were 16% higher in the exposed group (NOT "16% more patients had the event")
When does an odds ratio approximate the relative risk?
When the outcome is rare (under about 10%); for common outcomes, the OR exaggerates the effect
Why can relative measures mislead?
The same RR can mean very different ARRs and NNTs depending on the baseline risk
TRUE or FALSE: Body temperature in Celsius is ratio data.
FALSE. It's interval: equal spacing but no true zero (0°C doesn't mean "no temperature")
TRUE or FALSE: Mean ± SD should be reported for skewed data.
FALSE. Report median (IQR) for skewed data
A boxplot's median line sits close to the bottom of the box, with a long upper whisker. What is the skew?
Positive (right) skew; a few high values stretch the top
Length of hospital stay: mean 9 days, median 4 days. Which direction is the skew?
Positive (right) skew, since mean > median
2x2 practice: Drug group 20 of 200 had MI; placebo group 40 of 200 had MI. Find the risks.
Drug risk = 20/200 = 10%; placebo risk = 40/200 = 20%
2x2 practice (drug 10%, placebo 20%): find RR, RRR, ARR, NNT
RR = 0.5; RRR = 50%; ARR = 10%; NNT = 10
TRUE or FALSE: An RR of 1.0 means the treatment reduced risk.
FALSE. RR = 1.0 means no difference between groups
NNT calculates to 12.2. What do you report?
13 (always round NNT UP)
NNH calculates to 33.8. What do you report?
33 (always round NNH DOWN)
Incidence or prevalence: 50 new cases of type 2 diabetes this year among 1,000 adults without diabetes
Incidence (NEW cases among those at risk)
Incidence or prevalence: 120 of 1,000 adults currently have type 2 diabetes
Prevalence (ALL existing cases at a point in time)
Data type: number of seizures per month
Ratio (true zero: 0 seizures means none)
Data type: NYHA heart failure class I–IV
Ordinal (ranked, unequal spacing)
Data type: smoker vs non-smoker
Nominal (dichotomous)
Data type: serum potassium (mEq/L)
Ratio (true zero, equal spacing)
Why is it wrong to say 20°C is "twice as warm" as 10°C?
Celsius has no true zero (interval data), so ratios between values aren't meaningful
Best measure of center: blood type of 500 patients
Mode (nominal data)
Best measure of center: patient satisfaction rated 1–5
Median (ordinal data)
Best measure of center: normally distributed systolic BP
Mean
Best measure of center: hospital cost (heavily right-skewed)
Median (resists the long tail)
A trial reports a mean HbA1c drop of 0.8% with SD 0.2 vs SD 1.5. What's the difference?
SD 0.2: most patients dropped close to 0.8%. SD 1.5: patients varied a lot, so the mean describes few of them
Mean total cholesterol 200, SD 20, normal distribution. What range holds about 95% of patients?
160–240 (mean ± 2 SD)
Mean total cholesterol 200, SD 20, normal distribution. What range holds about 68% of patients?
180–220 (mean ± 1 SD)
Exam scores: mean 78, median 85. Which direction is the skew?
Negative (left) skew, since mean < median (a few very low scores)
Hospital cost: mean $8,400, median $5,100. What does this tell you?
Right (positive) skew; a few very expensive patients pull the mean up
A histogram shows two distinct peaks. What could that mean?
Two different populations may be combined in one dataset
On a boxplot, what does a dot beyond the whisker represent?
An outlier (more than 1.5 × IQR beyond the box)
A scatterplot shows a strong upward trend between coffee intake and heart rate. Does coffee cause higher heart rate?
Not proven; a scatterplot shows association, and correlation is not causation
Calculating incidence of FIRST MI: who is left out of the denominator?
Anyone who already had an MI; only people at risk count
Prevalence rose over 10 years while incidence fell. Most likely reason?
Patients with the disease are surviving longer
Does an RRR of 43% tell you how many patients need to be treated?
No. RRR is relative; only NNT (from ARR) tells you that
Why round NNT UP?
It avoids overstating the benefit (conservative)
Why round NNH DOWN?
Rounding up would understate the harm (conservative)
Interpret RR = 0.75
Treated patients had 25% lower risk of the outcome than controls
Interpret RR = 1.30
Treated patients had 30% higher risk of the outcome than controls
Interpret OR = 2.0
The odds of the outcome were twice as high in the exposed group
Interpret NNT = 25 in a 2-year trial
25 patients need to be treated for 2 years to prevent 1 event
Why can't you calculate RR in a case-control study?
The researchers chose how many cases and controls to enroll, so risk in the population can't be calculated; use OR