Professional Documents
Culture Documents
Measuring Placebo Effects
Measuring Placebo Effects
MEASURING PLACEBO
EFFECTS
Jeremy Howick
1. Introduction
medical research and practice, a placebo~ a
The term placebo is Latin for "I shall please." In
ve it is, or could be, the real treatment whtn
treatment that is capable of making people belie
is indistinguishable to a patient fromreal
in fact it is not that treatment. A sugar pill that
in clinical trials, placebo controls arc often
vitamin C, for example, could be a placebo. With
ents are compared. For example, when
used as a standard against which experimental treatm
patients a placebo ,md other patien ts the
evaluating a new drug, investigators can give some
rforms the placebo by a sufficient margin,
experimental drug. If the experimental drug outpe
le, doctors, outcome assessors, and daca
it is said to be effective. Patients (as well as, if feasib
means they don't know which treatment is
analysts) in these trials are often masked, which
Placebo use in clinical practice and clinical
the experimental drug and which is the placebo.
2009, Howick et al., 2013a). A great deal of
trials is both common and controversial (Howick,
are (G(lltzsche, 1994, Nunn, 2009), as well
philosophical controversy surrounds what placebos
ick, 2009). Ther e is also a dispute-the one
as whether placebos are ethical (Foddy, 2009, How
corre ct methodology that should be used to
I will concern myself in this chap ter-a bout the
measure placebo response.
ed that one-third of the benefit of all
Henry Knowles Beecher's early ( 1955) study claim
). This is still the most widely cited paper
treatments was due to placebo effects (Beecher, 1955
rful persist because of it. Yet more recent
in the field, and the beliefs that placebos are powe
Peter G(lltzsche in Denmark conclude that
studies conducted by Asbjom Hr6bjartsson and
bjartsson and G(lltzsche, 2001, Hr6bjartsSOn
placebos do not have important effects at all (Hr6
and O(1ltzsche, 2004a, Hr6bjartsson and O(1ltzsche,
2010). In this chapter I will review the
odological problems with these attempts.
attempts to quantify placebo effects and list the meth st
estimate is likely to be an overe imate,
To anticipate, I will argue that while Beecher's 1955 shows th3 t
more recen t studies unde restim ate place bo effects. A large body of evidence
the
ailments, especially pain.
placebo effects are very powerful for treating some
and the problems with it. The
I will proceed by first describing Beecher's 1955 study
rectly infers from the fact that someone
main objection to Beecher's study is that he incor
the recovery was due to the placebo (sec-
recovers after taking a placebo to the claim that
recen t studies conducted by Hrobj arrss,on
tion 2 ). I will then proceed to describe the more
ems with them. The main issue with Hrobd
~nd G¢tzsche and list the methodological probl
includes heterogeneous placebos an
Jartsson and G¢tzsche's analysis is that their study I I ebo
k· · ate compatible with powerfu Pac
cond.. tttons, ma mg their average placebo effect estim
134
MEASURING PLACEBO EFFECTS
f for certain conditions such as pain (section 3). Before concluding (in section 5), I
ef ect_s the evidence and mechanisms that explain why, despite the methodological dif-
descn 6e . . .
. I . . in estimatmg placebo effects, tt 1s reasonable to believe that placebos are effective
ftcu ues .
for creating pain (sec non 4) •
135
-. •-lorA •-• ~- ~ -- - - -- -
•. .>
JEREMY HOWICK
l. patients not treated at all (often people in these groups were placed on waiting lists)
2. patients given a placebo
J. patients given a "real" treatment
They identified 114 trials. Typical pill placebos were lactose pills, typical physical placebos were
procedures performed with the machine turned off (e.g., sham transcutane ous electrical nerve
stimulation), and typical psychological placebos were theoreticall y neutral discussion between
participant and dispenser. They then compared what happened to patients who received the
placebo with patients in the untreated groups. This method avoided the post hoc ergo procter hoc
fallacy because it included a comparison of patients whose ailment would have followed its natu-
ral history: the untreated group. They found that on average placebos only have small effects:
[There is] little evidence that placebos in general have powerful clinical effects. Pla-
cebos had no significant pooled effect on subjective or objective binary or continuous
objective outcomes. We found significant effects of placebo on continuous subjective
outcomes and for the treatment of pain but also bias related to larger effects in small
trials. The use of placebo outside the aegis of a controlled, properly designed clinical
trial cannot be recommend ed.
(Hr6bjartss on and G¢tzsche, 2001, p. 1599)
Even in cases where they found a significant placebo effect-not ably for treating pain-
Hr6bjartsson and G¢tzsche question whether the observed effects were actually placebo
effects. They raise a similar concern to Kienle and Kiene, noting that the outcomes in the pla-
cebo groups could have been due to a form of reporting bias, such as patients politely reporting
th ey had recovered when in
fact they did not actually experience any recovery. In their wo rd s:
~atients in an untreated group would know they were not being treated, and patients
tn a placebo group would think they were being treated. It is difficult to distinguish
between reporting bias d
. a
an a true enect o f placebo on subjective outcomes, smc · ea
pattent may tend to try to I h • .
h as occurred. The fact th P ease t e mvesttgato r and report improveme nt wh en none
l
at P ace bos h a d no significant effects on objective conttn· u·
ous outcomes suggests th t . b . l
. .
su6Jecttve outcomes. a reporting ias may have been a factor in the trta s wt.th
136
>
MEASURING PLACEBO EFFECTS
ter hoc fallacy, their conclusio ns nre questiom,ble becA use of scvern l problems with their
/JT'~ ., The mc1 in problems :u e: (I) tht' ir rev iew i11cluded heLcrogeneu us s l udi es that argur1 bly
~~ '
I Id not have been lumped together so their HVcrage rcsulLs fa il to rule out powerful placebo
.
137
w
r,f
JEREMY HOWIC K
the placebo concep t h:is proven notorio usly difficu lt to define .,d n equate ! I .
. . •
y. t 1s beyond I
of this chapte r to outlm e the definit ional proble ms- see Howic k ' 2016 for a · th1e "'-Oiie
•~
i: · · ns o f p I~ ·
•ce LJOS as mact1v · e or nonspe cific ., . review
t I1e comm on denntt1o . _ . . ,·,re c 1earl • - OI\I
.
can be active, c1t least with respec t to some mlmen ts, and their , I . ~ mistaken: I·acel. ever,
. f. I I . I d rnec 1an1sm f p
uos
. I 4) 'T' 0 actio
as specih c as t 1e actlLln o a p 1armac o og1ca rug (see section · ' 0 avoid h n can
· d G - I d 'd d
jbtzsc 1e ec1 e to adopt a practic al a t e defini . 1.
UI!
proble m, Hr6bJ..irtsson an 0
the~~roa_ch and charat '~af
placeo os "prnc tirn lly as an [any!] interv ention labeled as such in · • tr·,
Su ero, tuf •,' c11nrcal
k ' I ·h '· I · terize,,
· I .11 d
But 1t mm y nee s rcmar mg t 1.it t is approa c 1 1s untena ble.
. . I
. ·11· . , I lJ O ma tna o f some new a _. . . x,1111pfe, th• ppose, for l' . 1,1I.
.
someo ne repartee:I usmg pc111c1 111 ,1s a p ace nt1 6 tot,c as ,it
. Tt ·tt f b " if the d'd a treatrne111
for pneum onia. ne respon se w1 o course c no one would, and I
· 1y. " But t I11s
· I serious · reactio · n seems exact Iy to s h ow that we wy k we . would not takc
the tna
. . or w1t 11 so
ents about what can and canno t count as approp riate . _Ille concept
that involv es Judgm 1
H:!~~11late place.
bos and placeb o contro ls. Since it is unclea r wheth er the placeb os withi~r
effects m b Jansson and
G¢tzsc he's study were actual placeb os, their estima tes of placeb o ust e quest1.0
of ac . ned.
Worse , Hr6bja rtsson and G¢tzsc he go back on their alleged policy cepting a
le the dny treat.
ment labeled as a placeb o in the report of a clinica l trial. For examp I
t n~t assy e~c ud e ~tudies
where "it was very likely that the alleged placeb o had a clinica l benefi ociate with h
ent techni ques for postop erative pain)" (Hrobj artsson and G te
ritual c1lone (e.g., movem
os as the effect f "~_tzsche ,
2001 , p. 1595). Here they seem to sneak in a definit ion of placeb I b so rituals"
. not accept a bl e: ntua
. h 1s . 1 r1eastm . f: .
g or astmg are not p ace os. A similar proble .'
w h 1c
he's study may actual~ a;ises
becaus e the untrea ted groups within Hr6bja rtsson and G¢tzsc
and contac t with health care practit ioners. y ave
benefi ted from observ ation by
138
►
MEASURING l'LACEIIO EFI-El ~ S
. rhcric (Di Bl:,si et :ii., 200 1). This prescms a problem with Hro bj:1rrsson ;ind v0t zscl 1e's
enip,i · 'I I Ot1· ' I1 Ievel at Rccchcr. The uut corn cs in the Lm tre·tt cd
. , rh'1t is sltnt nr tll tic I Jccttun L ey
. ,
rev1e11 , .
effrct ~
ups 111 ight not be rhe resL'. t of natural l~istory, but instead be the re.~ulr nf Hawthor ne
grol I e effects of conrnct with empathe tic her1lth care prac titioners.
:in, r 1 b 1· ne
There is good reason to c ievc that the untrcr1ted groups did benefit fro m Hawthor
fou nd that the
a;. cs and the effects of caregiver attention . A recent syste mc1tic review
eceated groups within
cunrr Hr6bjartsson and G~rzsche's review experienc ed ·1 24% improvc-
. . '
wtth baselme (Krogsbo ll et a l. , 2009). There are two nlausible explanati o ns
n1ent Compared t
for the 24% improvem ent from baseline. First, it could be natural history of disease. Second,
rhe improvement is partly due to H awtho rne effects and effects of contact with h ealth care
a large
ractitioners. If it is the second- and this is my preferred explanati on since 24% is
then Hr6bjarts son
~(feet that is not likely to be explainab le solely in terms of natural history-
no t,
and G~tzsche have underesti mated the placebo effect. Because the untreated groups were
's review underesti mated placebo effects. This is
in fact, untreated , Hr6bjartsson and G¢tzsche
by subtracti ng outcome s
due ro simple arithmeti c, noting that the placebo effect is estimated
in the placebo effect with outcomes in the untreated groups.
139
JEREMY HOWICK
140
111111
for killing pain, Hr6bjartsson And Gy,tzsche are 111istHkcn that placebos have insignificr1nt
effects.
5. Conclusion
What cc1n we conclude about the power of placebos from these studies? The first thing is to
knowledge the fact that placebos won't work for everything. If someone has a leg amputated, '
::l\ing a sugar pill or believing it will grow back is unlikely to h<1ve any effect. In addition, ,I
~:any ailments-especially the most common ones people visit their doctors for, such as mild
to moderate pain, depression, anxiety, flu, colds, and so on-will go away on their own and
without any treatment (placebo or not). Beecher's early and widely cited systematic review of
placebo effects made the methodological error of failing to take natural history into account,
and it led to an overestitnation of placebo effects. However, Hr6bjartsson and G¢tzsche's con.-
clusion that placebos do not have much of an effect at all is equally problematic.Just as "real"
treatments are exceptionally powerful for treating some conditions (for example, antibiotics
for meningitis), so placebo treatments can be effective for treating some conditions, especially
mild to moderate pain and depression, anxiety, and smoking cessation, and they can generally
improve quality of life. By lumping all types of placebos for all conditions together, Hrobjarts ..
son and Gotzsche obscure important placebo effects for conditions like pain. For example,
the effects of placebo analgesia have been demonstrated empirically, and their mechanism
of action-endogenous opiates-has been established in numerous studies. Further focused
research is required to clarify the placebo concept and provide accurate estimates of placebo
effects for specific conditions.