OBJ - Probability & Diagnostic Tests

Question 1

Sensitivity

Answer 1

Define and interpret & calculate

SNOUT

– POSITIVE in those WITH disease

Sensitivity = a/(a+c)
Sensitivity = TP / (TP + FN)	Sensitive test if FN worse than FP

Sensitivity + False Negative = 100%

Using a sensitive test
• TP / (TP + FN) close to 100%
• Few false negatives = Most negative results are TRUE
• Patients with negative results likely do not
have disease
• Sensitive test with Negative result rules OUT disease (SNOUT)

Question 2

Specificity

Answer 2

Define and interpret & calculate
SPIN

– NEGATIVE in those WITHOUT disease

Specificity = d/(b+d)
Specificity = TN / (TN + FP)	Specific test if FP worse than FN

Specficity + False Positive = 100%

Using a specific test
• TN / (TN + FP) close to 100%
• Few false positives = Most positive results are TRUE
• Patients with positive results likely have disease
• SPecific test with Positive result rules IN disease (SPIN)

Question 3

Positive predictive value

Answer 3

Define and interpret & calculate

What your patient wants to know:
• “My test was positive—am I really sick?”
– Positive predictive value
TP / (TP + FP)
a / (a+b)		across row

HIGH PPV:
When disease is prevalent, most positives are true positives (few false positives)

LOW PPV:
When disease is rare, fewer positives are true positives (more false positives)

• As prevalence increases
– Positive predictive value increases
– Negative predictive value decreases

Prevalence is POSITIVELY associated with PPV

Question 4

Negative predictive value

Answer 4

Define and interpret & calculate

• “My test was negative—am I in the clear?”
– Negative predictive value
TN / (FN + TN)
d / (c+d) across row

Prevalence is NEGATIVELY associated with NPV

Question 5

Reference Ranges

Answer 5

DEF: A range of values intended to include 95% of persons assumed to be disease free
- Alternate terms: Normal range/Reference range

SOURCES OF REFERENCE RANGES
Subject based
– Requires historical data on individual
– Difficult to obtain

Population based
– Data from many individuals
– Traditional approach

DETERMINING REFERENCE RANGE
1. Normal distribution method: Mean ± 2 x standard deviation
Assumes normal distribution (often not true in practice)

2. Transformation method: Transform values (e.g. take logs) to make data more like a normal distribution, then use normal distribution method
3. Percentile method: works all the time
   Assumes no distributional form
   Sample size should be large enough so estimates of range will be reliable
   Disease risk not confined to “abnormal” values

PROBLEMS OF INTERPRETATION
- “Abnormal” tests are often “normal” on retesting “Regression Towards the Mean” -> Retesting persons with “high” lab values

• Routine test batteries flag many “abnormal” values (someone will always have something off if you test them enough)