Wednesday, May 4, 2011

Helping Doctors and patients Make Sense of Health Statistics


Gigerenzer et al

Psychological Science in the Public Interest –Volume 8, Number 2, pp 53 –

 

-          conditional probabilities – which include sensitivity and false positive rate

-          better to use natural frequencies

-          example

o   Conditional probabilities

 

 

 

1 woman

 

 

 

 

1% chance of breast cancer

 

 

 

99% no breast cancer

 

90% positive

 

10% negative

 

9% positive

 

91% negative

 

 

 

 

 

 

 

p(breast cancer | test positive) = (0.01 * 0.9)  / (0.01 * 0.09) + (0.99 * 0.09)

 

o   Natural frequency

 

 

 

 

1000 women

 

 

 

 

10 breast cancer

 

 

 

990 no breast cancer

 

9 positive

 

1 negative

 

89  positive

 

901negative

 

 

 

 

 

 

 

p(breast cancer | test positive) = 9 /( 9 + 89)

 

 

 

-          Many clinicians do not know probability person has disease given a positive screening test.

o   positive predictive value

 

Higher Survival Does Not Mean Longer Life

-          meaningless in making comparisons across groups of people that differ dramatically in how diagnosis is made.

-          5 year survival rate

o   number of patients diagnosed with disease still alive after 5 years / number of patients diagnosed.

-          annual mortality rate

o   number of people who die of disease in year / number of people in group

-          diagnosed appears in numerator and denominator of 5 year survival rate, but no where in mortality rate.

-          screening profoundly biases survival

o   affects timing of diagnosis (lead time bias)

o   affects nature of diagnosis by including people with non progressive cancers (over diagnosis bias)

-          survival rates can be increased by setting the time of diagnosis earlier, even if no life is prolonged / saved.

 

What is Statistical Literacy?

-          Learning to live with uncertainty – understand there is no certainty and no zero risk, but only risks that are more or less acceptable

-          Questions to ask of risk

o   risk of what

o   time frame?

o   how big

o   does it apply to me

o   screening tests

§  understand screening tests may have benefits and harms

§  understand screening tests can make two errors

·         false positives

·         false negatives

§  understand how to translate specificities, sensitivities and other conditional probabilities into natural frequencies

§  understand goal of screening is not simply early detection of disease, it is mortality reduction or improvement of life

o   treatments

§  understand that treatments typically have benefits and harms

§  understand size of benefit and harm

o   Questions about science behind numbers

§  quality of evidence

§  conflicts of interest

 

 

-          screening is itended to detect existing cancers at early stage à so it does not reduce risk of getting breast cancern

-          patients were asked what do you feel is the likelihood of you having a heart attack over next 12 months

o   likelihood depends on individual factors like

§  age / sex/ smoking / diabetes

o   patients risk estimates showed no correlation with these factors

 

-          relative risk reductions can cause exaggerated perceptions of treatment effects

 

 

-          do clinicians understand the number needed to treat, which is

o   number of patients that must be treated in order to save the life of one patient.

-          4 identical funding proposals  à cardiac rehab and breast cancer screening

o    relative risk reduction

o   absolute risk  reduction

o   absolute values from which the absolute risk reduction is computed

o   number needed to treat

 

§  relative risk reduction – seen as having greatest merit

-          geography is destiny

o   surgical treatments are often not based evidence

-          speciality is destiny

 

 

-          informed consent refers to ideal how doctors and patients interact.

 

 

-          asubtle way to induce the illusion of certainity – by analogies

o   war on cancer

-          prevalent use of relative risk

o   sometimes defended on basis that ratio measures are transportable to different populations with different baseline risks

§  this is main weakness as well, since they conceal underlying absolute risks

-          some citizens believe à 30% probability of rain tomorrow

o   will rain tomorrow 30% of time

o   in 30% of the area

o   on 30% of days for which announcement was made

è problem – no reference class is used

 

Prozac example

                psychaiatrist – thinking of all his patients

                patients thought of themselves alone

 

 

 

sensitivities / specificities à two reference classes

                patients with disease

                patients without disease

 

natural frequencies à same reference class

è all patients

 

 

 

 

 

No comments: