Reliability of diagnostic tests

Reliability of diagnostic tests

October 29, 2014

O. Paltiel and R. Calderon

Different combinations of high and low precision/reliability and validity

The challenge of diagnosis Appearances to the mind are of four

kinds. Things either are what they appear to be; or they neither are, nor appear to be, or they are, and do not

appear to be, or they are not, yet appear to be. Rightly to aim in all

these cases is the wise man’s task.Epictetus, 2nd century

= Reliability מהימנות Definition “the extent to which

repeated measurements of a stable phenomenon – by different people and instruments, at different times and places – get similar results”

= reproducibility or precision

Validity and reliability

A high reliability mean that in repeated measurements the results fall very close to each other; conversely, a low reliability means that they are scattered. Validity determines how close the mean of repeated measurements is to the true value. A low validity will produce more problems when interpreting results than a low reliability

Sources of Variation

Measurement Instrument- The means of making the

measurement Observer -The person making the

measurementBiologic Within individuals- Changes in people with

time and situation Among individuals- Biologic difference from

person to person

Sources of variation. The measurement of diastolic blood pressure.

Fletcher RH. Clinical Epidemiology. The Essentials.

Biological Variability

Clinical disagreement in interpreting diagnostic materials

Diagnostic material Clinical question Clinicians

Mammogram

Exercise electrocardiogram (ECG) Peripheral blood film

Should this woman be referred for surgical

assessment?

Is the ST-T response normal, borderline, or

abnormal?

Does this patient have iron deficiency?

Two radiologists examined the same 1214 mammograms

Two cardiologists

examined the same ECGs from 38 patients One cardiologist twice

examined the same ECGs from 38 patients

Two pathologists

examined the same films from 29 women with iron-deficiency

anemia One pathologist twice

examined the same films from 29 women with iron-deficiency

anemia

Clinical disagreement in interpreting diagnostic materials (cont.)

Diagnostic material Observed Agreementexpected by

chance

Kappa

Mammogram

Exerciseelectrocardiogram(ECG)

Peripheral bloodfilm

97%

57%

74%

69%

83%

91%

39%

33%

49%

51%

.67

.30

.61

.39

.65

The etiology of clinical disagreement

The Examiner1. Biologic variation in the senses2. The tendency to record inference rather than evidence3. Ensnarement by diagnostic classification schemes4. Entrapment by prior expectation5. Simple incompetencyThe examined1. Biologic variation in the system being examined2. Effects of illness and medications3. Memory and rumination4. Toss-upsThe examination1. Disruptive environments for the examination2. Disruptive interactions between examiners and examined3. Incorrect function or use of diagnostic tools

Alvan Feinstein

“To advance art and science in clinical examination, the equipment a clinician most needs to improve is himself”

Effect of lack of sleep on clinical acumen

RestedFatigued

Mean hours of sleep prior to testing7.01.8

Mean number of symptoms Physiologic (weakness, nausea, etc.) Perceptual (trouble focusing eyes, etc.) Cognitive (difficulty concentrating, etc.)

1.10.11.4

5.41.44.6

Mean number of errors on ECG interpretation

5.29.6

Observers are biased: eg “context bias”

.

Observers are biased Comparison of fetal heart rate by

auscultation with rates obtained by fetal monitoring:

When true fetal heart rate was in the

normal range, rates by auscultation evenly distributed around true value (ie random error only).

When true fetal heart rate abnormally high or low rates by auscultation biased toward normal.

Day BMJ 1968;4:422

A measure of agreement between two clinical assessments

Observer I Positive Negative

Observer Positive a b a+b

II Negative c d c+d

a+c d+b

Definitions Ao = a+d= observed agreement N=a+b+c+d= Maximum potential agreement Ae= (a+b)/n x (a+c)/n x n + (c+d)/n x (b+d)/n x n = Expected agreement by chance (assuming independence)

Developing a useful index of clinical agreement

The Kappa statistic

K = (Ao – Ae)/(N- Ae)=

= actual agreement beyond chance

potential agreement beyond chance

Example I Observer I

Positive Negative

Observer Positive 46 10 56

II Negative 12 32 44

58 42 100Ao=78, Ae= (58X56)/100+(42X44)/100=51K=(78-51)/(100-51)=0.55

Example II Observer I

Positive Negative

Observer Positive 82 8 90

II Negative 5 5 10 87 13 100Ao=87, Ae=79.6K=(87-79.6)/(100-79.6)=0.36

Properties of Kappa Can be negative Cannot be larger than 1 Excellent agreement: Kappa > .75 Good agreement: Kappa [0.40-

0.75] Poor agreement: Kappa <.4

Continuous traits For continuous traits e.g

cholesterol a correlation coefficient can be

used