At 10:20 26/06/98 +0000, you wrote:
>I have some students (midwives studying a MSc module in epidemiology)
>who are having great difficulty getting to grips with the difference
>between relative and absolute risk, in particular the application of
>such concepts in health care. Does anyone know of any references I can
>guide them to that will help them??
I have collated resources sent to me. I am reluctant to send this as a
file attachment. Apologies for length. References/resources are collated
in a separate email.
Many thanks for all those
who responded.
Jane Sandall
Reader in midwifery
City University
1) Get them to write down the chances they think they have of
- being treated for depression in the next 10 years
- getting pre-eclampsia in their next pregnancy
- getting breast cancer
- getting cervical cancer
- getting lung cancer
- getting heart disease
This can provike an interesting discussion. Inevitably, risk of
depression is underestimated (it is the commonest reason for medical
treatment in the 25-40 age group in many studies!) while risk of lung ca
and especially cervical ca are greatly overestimated.
This leads to idea 1: your risk is our rate
My risk of a disease is the rate at which it occurs in people like me
Now, idea no 2. Get them to write down the effect that smoking would
have
on their risk of each of these diseases. Once again, discuss the
estimates - are they in line with the evidence. Focus in on lung cancer
and heart disease. Which is worth more
- a nine-fold reduction in your risk of lung cancer (which you're not
likely to get anyhow)
- a 3-fold reduction of your risk of heart disease (which is very
likely)
_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
_/_/_/ _/_/ _/_/_/ _/ Ronan M Conroy
_/ _/ _/ _/ _/ _/ Lecturer in Biostatistics
_/_/_/ _/ _/_/_/ _/ Royal College of Surgeons
_/ _/ _/ _/ _/ Dublin 2, Ireland
_/ _/ _/_/ _/_/_/ _/ voice +353 1 402 2431
[log in to unmask] fax +353 1 402 2329
_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
2) Maybe the best way of explaining Relative Risk is by way of an
example.
Lets say we want to measure the effect of a drug (call it X) on nausea
during pregnancy.
Of 40 women taking X, 8 experienced nausea.
Of 100 women not taking X, 40 experienced nausea.
The incidence of nausea in the X group is 8/40 = 20%
The incidence of nausea in the "no X" group is 40/100 = 40%.
The Relative Risk is defined as
incidence rate among exposed = 20%/40% = 1/2
incidence rate among nonexposed
We interpret this as: those receiving drug X are half as likely to
experience nausea as those not receiving drug X.
Absolute Risk is the same thing as incidence (the absolute risk of
experiencing nausea whilst on drug X is 20%).
Sean McGuigan
Boehringer Ingelheim
3)Relative risk (RR): likelihood of disease in exposed individuals
relative to those who are non-exposed.
attributable risk (AR): absolute effect of the exposure or the excess
risk of disease.
Risk of mortality in a cohort of British male physicians
Annual mortality rate per 100,000
Lung cancer Coronary Heart Disease
Cigarette smokers 140 669
Non smokers 10 413
relative risk 14.0 1.6
attributable risk 130/105/year 256/105/year
_______________________________________________
Martin Frischer
Senior Lecturer
Department of Medicines Management
Keele University
Staffordshire ST5 5BG
tel: 01782 583 568
fax: 01782 713 586
e-mail: [log in to unmask]
______________________________________________
4) maybe this also helps:
Relative Risk = is the risk of contracting a condition (disease) amongst
those exposed to a certain risk factor compared to the non-exposed.
Formula = Incidence in the exposed
..........................................................
Incidence in the non-exposed
Relative risk tells us how many times more (less) likely someone exposed
to a risk factor is to develop a disease (condition). Hence this is a
relative indicator of risk between the two groups and does not say
anything about absolute risk.
Absolute Risk = is the excess risk caused from being exposed to a risk
factor - over and above the risk faced by anybody not exposed to the
risk factor under investigation.
Formula = Incidence in the exposed - incidence in the
non-exposed
Hence the absolute risk tells us the excess number of cases that are
attributable to exposure.
Judith Strobl, Samuel Ghebrehewet
5)Enclose some notes on RR/AR I have accumulated, with some
references and web sites. I have used an attatchment because
there is a table, that won't work in pure e-mail, but in case
anyone has problems with attachments, I include the text below
Jane's request as an appendix, but you will have to re-construct the
table (or ask me for hard copy if required)
Regards Alan
+++++++++++++++++++++++++++++++++++++++++
**Absolute and relative risk
I must admit that I find reading statistics incredibly hard, and the
only way I can learn anything is to do a worked example, or construct
my own while reading. So, writing this section was quite a learning
experience for me. I suggest check my calculations.
**Absolute and relative risk
Let us take a simplified hypothetical example....
Suppose you are drawing up a protocol for the primary prevention of
cardiac events (fatal and non-fatal proven myocardial infarction) in
middle aged men.
You have two groups, of 2000 patients each:
*A high risk group, of smokers (five or more cigarettes per day for
over 5 years); hypertensive (systolic over 160 mm Hg) and
hyperlipidaeic (total cholesterol over 6.5mm).
*A low risk group of non-smokers (non-smoker for at least five years);
borderline normotensive (systolic 140-160mm Hg, diastolic under 95
mm Hg); total cholesterol. under 5.8mm.
Within each group, subjects are observed for one year and randomised
to either:
* i. No active treatment
* ii. Atenolol 50mg daily.
Both groups have been chosen to exclude patients with absolute
contra-indication to beta-blockade.
So, there are four groups, of 1000 each:
i. High risk, beta-blocked.
ii. High risk, not beta-blocked.
iii. Low risk, beta-blocked.
iv. Low risk, not beta-blocked.
Okay, so there are faults in the methodology, and if anyone wants
to comment on them fine, but this is a simple example, and I am
going to "assume" some reasonable results. By the way, the nice
convenient numbers aren't a fudge, as we will be using rates, rather
than absolute numbers, and the size of the four groups is irrelevant,
as long as all are large enough to reach statistical significance.
Table below show the results and the formula for calculating relative
risk reduction, absolute risk reduction and number needed to treat
[to prevent one adverse event].
Control group
Treatment group
No. of cardiac events Rate No. of cardiac events
Rate
High risk 20 (n=1000) 0.02 14 (n=1000) 0.014
Low risk 10 (n=1000) 0.01 7 (n=1000) 0.007
Relative risk Absolute risk reduction
reduction (RRR) (ARR)
High risk 0.3 0.006
Low risk 0.3 0.003
Number needed to treat (NNT)
High risk 167
Low risk 333
*Relative risk (RR):
This is defined as:
(the probability of an event in the active group) divided by ( the
probability of the event in the control group).
For the high risk group this is 0.014/0.02 and for the low risk it is
0.007/ 0.01, i.e 0.7 for both. A benefical treatment will give an
RR less than one, so it looks like beta-blockers offer benefit to
both groups. In other words, in both groups those who are
beta-blocked have only 70% of the risk of a cardiac event of those
who are not beta-blocked. As a rough rule of thumb, a RR of >50% is
regarded as definitely clinically significant and one between 25 and
50% probably clinically significant.
**Relative Risk Reduction (RRR):
This is calculated by subtracting the RR from one. i.e RRR of 0 = the
active treatment is neither of benefit nor harm. RRR can also be
expressed as: the absolute risk divided by the probability of an
event in the control group
.However you calculate it, both groups show a RRR of 30%,i.e
beta-blockade reduces the risk of a cardiac event by 30%, the
corollary of the discussion under relative risk.
For some diseases, the benefit a treatment offers remains constant
even if the risk in the untreated group is not constant across
different patient populations. But, it is often important to consider
how frequent the risk of adverse effects is in untreated patients
before recommending treatment, because, although the relative risk
reduction is alwaysthe same (this treatment prevents 30% of the
adverse events that would happen in the untreated group), the
absolute benefit will varywith the base line risk in the different
groups.
Or, if we decide to base our reccommendations on the RR calculation,
this would be okay if the risk of cardiac events was the same in both
groups. But, we know that one group, because of their smoking,
lipids and blood pressure are more likely to suffer cardiac events
than the other. In order to take this into account, we need to
calculatethe absolute risk reduction.
**Absolute risk reduction (ARR):
This is the difference between the event rates in the two
groups,where the adverse event rate is less in the intervention group,
this suggests the intervention is beneficial.We now find that the
high risk group has a higher ARR than the low risk, and this reflects
the fact that in this group beta-blockade is more likely to prevent a
cardiac event because cardiac events are more common.
However, the ARR, as decimal fraction is not easy to grasp, so
often its inverse is quoted, and this is the same as the number
needed to treat to prevent one event. So, assumimg our follow up is
one year, we have to treat 333 low risk patients for a year to stop
one having a cardiac event, but we only need to treat 167 high risk
patients to achieve the same. We can now introduce some sort of
benefit / side effects balance: is it worth exposing 166 patients to
the side effects of beta-blockers in order to save one patient from a
cardiac event?
**Odds ratio:
Defn:
(the odds of an event in the treatment group) divided by (the odds of
an event in the control group)
Although useful in epidemiology, the odds ratio is rarely used
clinically.
The above example looks at reducing risk but the technique can be
applied to the risk of producing adverse effects. A very good example
of this was provided by the recent "scare" over certain
contraceptives. In this case (and I am approximating my figures) using
these drugs increased the relative risk of a thrombo-embolic event
fairly dramatically (??50-100%), but the absolute risk for both pill
and non-pill users remained low (say an increase from 1/20,000 to
1/40,000) and still lower than the risk of such an event during a
pregnancy ( say, 1/10,000).
Risks, comparison of:
For comparison f two groups with respect to risk of some events.
Group 1 Group2 Total
Outcome present a b a + b
Outcome absent c d c + d
Total a + c b + d n
Relative Risk, RR , = a / (a +c)
b / (b + d)
For the null hypothesis, RR = 1
Confidence intervals can be calculated as:
SE (Loge RR) = 1 - 1 + 1 - 1
a a + c b b
+ d
One small correction, in the table, I have ommitted to include the
absolute number of cardiac events in the low risk treatment group,
which should be 7.
Alan O'Rourke
Information Officer
Institute of General Practice
Community Science Centre
Northern General Hospital Sheffield S5 7AU
Tel: 0114 2714302 Fax: 0114 2422136
E-mail: [log in to unmask]
5) I am involved in teaching this concept to clinicians, students,
and patients (mainly seniors) around the province. By the end of each
session almost all of them, seniors included, understand the concepts
and
are able to figure out and apply the concepts of absolute and relative
risk
reduction.
The approach I take is the following:
A) don't give them any formulas, work with them to understand the
concept
with some real life examples
B) fool them with numbers first - for instance, I ask clinicians and
patients the following question
Imagine that you just found out you have a risk factor for
cardiovascular
disease (e.g., high blood pressure or high cholesterol). A drug that
will
treat this risk factor is available and it has no side effects and its
cost
is covered by a plan. Please consider the following three scenarios.
Would
you be willing to take this drug every day for the next five years if it
had been shown in a clinical trial that:
1) patients treated with this cholesterol pill had been shown to have
33%
fewer heart attacks than the non-treated patients; or if
2) it was found that 2% of the patients who took this cholesterol pill
had
a heart attack, compared to 3% who did not take this pill - a difference
of
1%; or if
3) in 100 patients who took this cholesterol pill for five years the
medicine would prevent one of the 100 from having a heart attack. There
is
no way of knowing in advance which person that might be?
About 80% will say yes to number 1 but only 20-30% will say yes to
number 2
or 3
We then tell them that if you didn't make the same decision for all
three
scenarios you just got caught by statistics because the three scenarios
were from the same trial presented to you in three different ways.
C) Fool them again by asking who would NOT like to get the following
letter
from your bank
Dear customer
It is our pleasure to announce that effective immediately our bank is
reducing mortgages by 30%
Almost no one will say they wouldn't want this letter - then ask how
many
of them actually have mortgages - for those who don't have mortgages
this
letter means they have saved 30% of NOTHING.
D) What we eventually get our seniors to say is that when you hear
numbers
like aspirin reduces heart attacks by 30%, cholesterol drugs decrease
the
chance of heart attacks by 35% always ask 30% OF WHAT!!
E) We give them the following article
To help show consumers how to interpret or deconstruct a recent Bristol
Meyers Squibb ad, I and a Vancouver writer (Jim Boothroyd) wrote an
article in the Summer edition of magazine call AdBusters - a Vancouver
magazine with international recognition whose aim it is to expose ads
for
what they are and to create awareness of the power of advertising. I
believe the title of the article is "Drug Ad Hard to Swallow".
The web site address for AdBusters is http://www.adbusters.org but I
don't
think you can get access to this specific article so I have attached the
text below. Hope you fnd it interesting. Any comments would be
appreciated.
THE FOLLOWING IS WHAT IS IN THE HEART ATTACK AD
As you know, drug companies are now starting to directly market
prescription drugs to the public. A recent ad depecting Darryl Sittler
(a
Canadian sports icon) was printed as a full page ad from Bristol-Myers
Squibb in the Vancouver Sun stating that:
"Darryl Sittler knows about the risks of heart atacks ... so should
you."
This is followed by a picture of Darryl and his family.
"Darryl lost his father to a a heart attack.
He believes in knowing your cholesterol risk.
One out of three first heart attack victims die not knowing their risk
One particular medication, with a good diet and lifestyle, can reduce
the
risk of first heart attacks by 31% and second heart attacks by 62%
Call know if not for you but your family"
This is then followed by a suggestion to phone a 1-888 number.
WHAT FOLLOWS IS ROUGHLY THE TEXT OF OUR ARTICLE
Perhaps you've noticed that pharmaceutical companies are now pitching
their
prescriptions directly to consumers. "This is the information age and
more
information empowers patients to be able to have more meaningful
conversations with their doctors about cures and treatments", explains
Alan
Holmer, president of the Pharmaceutical Manufacturing Association.
Really? How then to explain the ad on this page? Darryl Sittler knows
about
the risks of heart attack. - so should you! This full-page plug for a
cholesterol-lowering drug called Pravachol ran in Canadian daily
newspapers
last winter. At first glance, one might mistake it for a public-service
announcement. No product is named and the reader's eye is drawn to a
clip-out coupon for a free Heart Attack Prevention Line. And doesn't
that
logo recall the old heart-shaped emblem of the Canadian Heart & Stroke
Foundation? Looking closer, one notices the reference to one medication
and, in the finest print, the name Bristol-Myers Squibb, the fifth
largest
pharmaceutical company in the world.
Darryl lost his father to a heart attack. He believes in knowing your
cholesterol risk. One out of three first heart attack victims die not
knowing their risk.
As true Canadians know - Darryl Sittler was captain of the Toronto Maple
Leafs and scored 484 goals for the Leafs from 1970 to 1981. Sittler was
fighting fit, and probably still has a strong ticker. So if he's
worried,
we should be terrified, right?
One particular medication, with a good diet and lifestyle, can reduce
the
risk of first heart attacks by 31%. But what does this mean? That if you
take this medication, eat your greens and walk to work, you're a third
less
likely to have a first heart attack?
This 31% comes from a clinical study published in the New England
Journal
of Medicine in 1995. The study looked at a group of 6,600 men from
western
Scotland. The average age of the men was 55, all had high cholesterol
and
45% were smokers. Half were given a placebo, half pravastatin,
(Pravachol.)
Those who took the placebo had a 7.9% chance of a first heart attack,
while
those who took the drug every day for five years had a 5.5% chance.
The difference is 2.4%, much lower than the 31% reported in the Sittler
ad.
Why the disparity? The 31% comes from dividing the difference in risk
between the
placebo group and drug group, 2.4% (7.9%-5.5%) by the risk in the
placebo
group (2.4%/7.9% = 31%).
An analogy might help. Your bank announces it's slashing mortgages by
30%.
If you have a mortgage worth $1 million, a cut of 30 % means your bank
has
forgiven $300,000 of your debt - good news. But if you have just $1,000
left on your mortgage, a 30% cut means you save just $300. And if you
have
no mortgage, then you save nothing.
The same applies for the Sittler ad. How much you would benefit from
Pravachol depends on what your risk of heart attack is to begin with.
If you're a 55-year-old woman with high cholesterol but no other risk
factors, your chance of a heart attack in the next five years is about 3
percent. Taking
Pravachol every day for five years may reduce your risk by 31% of 3%, or
1%
- The risk would go down from 3 down to 2% or a one in 100 chance of
benefitting from this medication. However, this drug has not been tested
in
women with heart disease to even really know the answer.
Given the benefit in the clinical study was really 2.4%, one might
question
the exclamation marks in the ad. Unfortunately it is not just consumers
who
are taken in: doctors, nurses and pharmacists are often confused by the
same sorts of statistics.
Free heart attack risk evaluation. Speak to a qualified nurse. Jim
Boothroyd spoke to Sheila, one of about 20 nurses answering calls for
the
heart attack prevention line at Bristol-Myers Squibb. Sheila said she
wouldn't mention the medication or the study - unless the caller fell
into
a high-risk category.
"Many callers are surprised that they are at risk that is why I think
that
it's important to have a line such as this one," Sheila said. "We need
to
promote health from awareness." But Bristol-Myers Squibb might have
raised
our awareness further if it had stated: "For Scottish males with high
cholesterol, half of whom smoke, a medication which costs approximately
$4,000 for five years of therapy, with a good diet and lifestyle, can
reduce the risk of a first heart attack by just over 2%." But that kind
of
copy wouldn't keep Sheila and her colleagues busy answering the phones.
F) Finally, at least for clinicians, we give them the following article
which gives them a tool which helps them communicate CHD risks and
benefits
to their patients - Primary prevention of heart disease and stroke:a
simplified approach to estimating risk of events and making drug
treatment
decisions. Can Med Assoc J 1997;157:422-8
Hope some of this info is useful and if you have any questions please
let
me know.
James McCormack, Pharm.D.
Associate Professor
Clinical Division Chair
Faculty of Pharmaceutical Sciences
c/o Pharmacy Department
St. Paul's Hospital
1081 Burrard St.
Vancouver, B.C.
Canada
V6Z 1Y6
604-631-5150 Fax 604-631-5154
6) Here is an analogy that I once read. I use it a lot with my
patients.
In my home state there is a lottery. The odds of winning, if I buy one
ticket, are 1 in 7.5 million. If I buy two tickets, I have doubled my
chances. I have relatively increased my chances 100%. However, I have
increased my real chance of becoming wealthy to 2 in 7.5 million. My
absolute
increase in odds of winning is only 1 in 7.5 million. Based on the
relative
"risk" increase, I should buy two tickets. Based on the absolute
change, the
smart move is not to buy any tickets.
Hope this helps.
Cheers and Best Wishes,
Brian
...................................................................
Brian Budenholzer, MD
Director, Clinical Enhancement & Development
Group Health Northwest
CC16
PO Box 204
Spokane, WA 99210-0204
USA
[log in to unmask]
509/ 838-9100 X 7393
fax: 509/ 458-0368
7)I teach a distance-education survey-level epidemiology course entitled
"Nursing and the Health of Communities" in which these concepts are
included.Application is then explored in a course workbook and
individual tutoring
given to the distance students. What seems to help them grasp the
concepts
is asking them to consider who would be most interested in these
statistics.
Faced with a practical example, most seem to appreciate their value, for
example:
If they were hired to plan or run a new nursing clinic, how would they
anticipate potential caseload to estimate required staffing and supply
levels: they would need prevalence rates of their community;
How would they recognize trends in occurrence of pertinent conditions:
they
would need to watch incidence rates;
How would they know which people to counsel or which interventions might
improve outcomes for their patients: they could examine relative risk to
identify risk markers or risk factors;
How would they know which factors promise the "biggest bang for the
buck" so
that they could target the most cost-effective risk factor(s): they need
to
know risk difference to see size of potential impact in absolute terms.
Do you cover SMR in your course? If so, what kind of nursing examples do
you
use for that? I've used comparison of nosocomial infections among
surgical
patients in one hospital versus another to illustrate this method of
adjusting for confounding (by case-mix in that example), then test their
understanding with a homework problem on interpretation of regional
hysterectomy rates (in which age distribution is the confounding
variable).
The formulae, the often arbitrary nature of the reference population,
and
the concept of SMR cause difficulty for many of my students.
David Birnbaum, PhD, MPH
Clinical Assistant Professor
Dept. of Health Care & Epidemiology
University of British Columbia, Canada
8)I don't know if this will help or not but......
Absolute risk is the likelihood of something happening. If you drive you
stand a 1 in 200 chance of crashing in any one year. The relative risk
is an increase or decrease in risk caused by some factor. Driving while
under the influence will increase your absolute risk of crashing. Say
the absolute risk is now 1 in 50. In other words you are four times more
likely to have an accident if you drive while drunk vs sober. The
relative risk is four or 400% cf sober drivers. If you drive a volvo the
absolute risk may be 1 in 150, this gives a relative risk vs all cars
of 200/150 = 1.33, or 33% more likely to crash.
I apologise to all safe Volvo drivers.
[log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
