sensitivity and specificity part ii – computations and examples

12
Sensitivity and Specificity Part II – Computations and Examples

Upload: linda-houston

Post on 17-Dec-2015

225 views

Category:

Documents


1 download

TRANSCRIPT

Sensitivity and SpecificityPart II – Computations and Examples

This video is designed to accompany

pages 81-94

in

Making Sense of UncertaintyActivities for Teaching Statistical

ReasoningVan-Griner Publishing Company

ImPACT Example

Concern about concussions in young athletes is at an all-time high. A variety of screening tests are available for assessing concussion status, some are to be used on the field; others take more time. One such test is the “ImPACT” test, which stands for “Immediate Post-Concussion Assessment and Cognitive Testing.”

How well does ImPACT perform as a screening test?

ImPACT Example

Rule: Beyond a certain level on the ImPACT scale you are said to be

concussed.Actual Status

 

ImPACT Prediction Athlete was, in

reality, not concussed

Athlete was, in reality,

concussed TotalsImPACT said “not

concussed”59 13 72 

ImPACT said “concussed”

7 59  66

Totals 66 72 138 

“Sensitivity and specificity of the ImPACT Test Battery for concussion in athletes.” Archives of Clinical Neuropsychology, 2005

FPR = 7/66 = 0.11 or 11%FNR = 13/72 = 0.18 or 18%

Specificity = 89%Sensitivity = 82%

Beck Inventory Example

The original BDI, first published in 1961, consisted of twenty-one questions about how the subject has been feeling in the last week. A typical question is below.

Choose the most appropriate response (0) I do not feel sad. (1) I feel sad. (2) I am sad all the time and I can't snap out of it. (3) I am so sad or unhappy that I can't stand it.

How well does Beck perform as a screening test?

Beck Inventory Example

Rule: If your total score on the Beck Inventory is 10 or greater then you

will be categorized as clinically depressed.

Actual Status as determined by

Diagnostic and Statistical Manual of Mental Disorders IV

 

Beck Inventory Indication

 Patient was, in reality, not depressed

Patient was, in reality,

depressed TotalsBeck said “not

depressed”66 5 71

Beck said “depressed”

12 12  24

Totals 78 17 95 

“Sensitivity and Specificity of Depression Questionnaires in a College-Age Sample.” Journal of Genetic Psychology, 2008, 169(3), 281–288

FPR = 12/78 = 0.14 or 14%FNR = 5/17 = 0.29 or 29%

Specificity = 85%Sensitivity = 71%

Field Sobriety Testing

There are three common testing procedures implemented during an FST: horizontal gaze nystagmus (HGN), one-leg stand (OLS), and walk and turn (WAT).

Data from 296 subjects who participated in the NHTSA’s 1998 San Diego field sobriety test validation study, ordered by total scores on these three FST procedures, are available .

Let’s set a cutoff for the field rule and assess how well the rule is working as a screening test for intoxication.

Rules

For The Test: assume that a 4 or above on the FST will tag a participant as drunk.

Concerning The Truth: assume that a BAC of 0.04% or greater means a participant is legally drunk. There were 267 study participants with BACs of 0.04% or greater.

Task at Hand

Using Total FST >=4 to Designate “Drunk”

FST Decision

Actual BACTotal

s< 0.04 (Sober)

>= 0.04 (Drunk)

Sober

Drunk

Totals 29 267 296

Fill out this table!

Field Sobriety TestingCase HGN OLS WAT Total FST

Actual BAC

229 0 0 1 1 0254 0 0 1 1 0.0266   1   1 0.067

142 2 0 0 2 0.005217 2 0 0 2 0.03191 2     2 0.034182 2 0 0 2 0.038109 2     2 0.04259 2 0 0 2 0.04199 0 1 1 2 0.048113 2 0 0 2 0.0522 2     2 0.0667 2   1 3 0.022

145 0 3 0 3 0.0353 2 0 1 3 0.03215 2 0 1 3 0.04

287 2 0 1 3 0.0489 2 1 0 3 0.05

123 0 1 2 3 0.05258 2 1 0 3 0.05335 2 0 2 4 011 2 1 1 4 0.01

247 4     4 0.0166 2 0 2 4 0.02

294 2 0 2 4 0.02231 2 0 2 4 0.0374 2 1 1 4 0.04

214 2 1 1 4 0.0458 4     4 0.0514 2 1 1 4 0.05834 4 0 0 4 0.05812 2 0 2 4 0.06

211 2 1 1 4 0.06232 4 0 0 4 0.06293 2 2 0 4 0.06130 2 1 1 4 0.07297 4     4 0.08271 2 0 2 4 0.1119   2 2 4 0.121

Using Total FST >=4 to Designate “Drunk”

FST Decision

Actual BACTotals

< 0.04 >= 0.04

Sober

Drunk

Totals 29 267 296

209 11

276

20 256

FPR = 20/29 or 69%Specificity = 9/29 or 31%

FNR = 11/267 or 4%Sensitivity = 256/267 or 96%

What If?

Change the rule: assume that a 2 or above on the FST will tag a participant as drunk.

Changes that occur: FPR will surely go up and FNR will surely go down.

So the point is: how well a screening test performs is directly related to the cutoff it uses to identify a “positive.”

One-Sentence Reflection

Simple fractions are used to compute sensitivity and specificity in many situations where both test results and the truth are arrayed in 2x2 tables.